Advances in 3D Geo-Information Sciences

Lecture Notes in Geoinformation and Cartography Series Editors: William Cartwright, Georg Gartner, Liqiu Meng, Michael ...

Author: Thomas H. Kolbe | Gerhard König | Claus Nagel

125 downloads 4304 Views 8MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Lecture Notes in Geoinformation and Cartography Series Editors: William Cartwright, Georg Gartner, Liqiu Meng, Michael P. Peterson

For further volumes: http://www.springer.com/series/7418

.

Thomas H. Kolbe (Eds.)

l

Gerhard Ko¨nig

l

Claus Nagel

Advances in 3D Geo-Information Sciences

Editors Thomas H. Kolbe Technische Universita¨t Berlin Fachgebiet Methodik der Geoinformationstechnik Sekr. H 12 Strabe des 17. Juni 135 10623 Berlin Germany [email protected]

Gerhard Ko¨nig Technische Universita¨t Berlin Fachgebiet Methodik der Geoinformationstechnik Sekr. H 12 Straße des 17. Juni 135 10623 Berlin Germany [email protected]

Claus Nagel Technische Universita¨t Berlin Fachgebiet Methodik der Geoinformationstechnik Sekr. H 12 Strabe des 17. Juni 135 10623 Berlin Germany [email protected]

ISSN 1863-2246 e-ISSN 1863-2351 ISBN 978-3-642-12669-7 e-ISBN 978-3-642-12670-3 DOI 10.1007/978-3-642-12670-3 Springer Heidelberg Dordrecht London New York Library of Congress Control Number: 2011921705 # Springer-Verlag Berlin Heidelberg 2011 This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: deblik, Berlin Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

Preface

The shape, extent, and location of spatial objects and environmental phenomena as well as the spatial distribution of physical and environmental characteristics are increasingly being described using three-dimensional (3D) geospatial representations today. In general, 3D modeling does not only affect the dimensionality of the spatial representation but also introduces the thematic structuring and decomposition of objects and phenomena along an additional – often the vertical – axis, leading to models that are typically much higher structured than 2D models. However, building modeling, urban and landscape modeling, modeling of the lithosphere and topography of the Earth, and Earth system modeling all have different requirements on the specific spatial representation and have brought forward a multitude of different 3D modeling frameworks and paradigms. With the additional consideration of temporal aspects another representational dimension is reflected, often being referred to as 4D modeling. One of the most active research fields has been recently the integration of spatial and spatio-temporal aspects together with thematic information from diverse application domains. Semantic 3D modeling addresses the thematic attributation and thematic interrelationships according to multiple domains. Representing the 3D/4D spatio-temporal properties along with the thematic aspects of the different domains is sometimes also called n-dimensional (nD) modeling. The complex structuring of geospatial information according both to spatial representations and semantic representations raises issues of spatio-semantic coherence that the scientific community has started to investigate only recently. Three-dimensional, four-dimensional, and n-dimensional models require efficient methods for the storage, retrieval, analysis, and visualization. Furthermore, standards are required that ensure the lossless exchange of information between the distributed components of spatial data infrastructures. New application domains require the development of new concepts for the representation of 3D space and three-dimensional spatial properties of real world entities and phenomena. In order to present the current state of the art in 3D geoinformation science and to further discuss these research topics the International Conference on 3D GeoInformation 2010 was held in Berlin, Germany. It is the successor of the four workshops on 3D GeoInformation that were carried out in the years 2009

v

vi

Preface

(Ghent, Belgium), 2008 (Seoul, South Korea), 2007 (Delft, The Netherlands), and 2006 (Kuala Lumpur, Malaysia). The 3D GeoInfo 2010 conference was conducted under the auspices of Working Group IV/8 of the International Society for Photogrammetry and Remote Sensing (ISPRS), the Open Geospatial Consortium (OGC), European Spatial Data Research (EuroSDR), the German Society for Photogrammetry, Remote Sensing, and Geoinformation (DGPF), and Berlin University of Technology. With more than 150 participants the 3D GeoInfo Conference 2010 offered an extensive interdisciplinary forum for international researchers from academia, industry, and government in the field of 3D geoinformation. In two keynote talks given by Maik Thomas from the GeoForschungsZentrum Potsdam (GFZ) and Ron Lake from Galdos Inc., 30 oral and 19 poster presentations, and an industry exhibition many different aspects of 3D geoinformation science were addressed and discussed. This book contains selected papers of highest quality that were presented at the conference. They have gone through a rigorous double-blind review process and were examined by three members of the program committee each. Afterwards the papers have been edited by the authors again in order to reflect the comments and suggestions of the reviewers. All other conference papers and extended abstracts that were accepted for oral and poster presentation are provided in a separate proceedings volume published by ISPRS within their International Archives of Photogrammetry and Remote Sensing (IAPRS), Vol. XXXVIII-4, Part W/15. Such a big event could not be realised without the help of many people. We thank the organising committee chaired by Bernd Stary (in alphabetical order): Thomas Becker, Andreas Fuls, Javier Herreruela, Robert Kaden, Andreas Kru¨ger, Rosemarie Kunkel, Hartmut Lehmann, Alexandra Lorenz, Sven Weisbrich. We owe special gratitude to Herbert Krauß from DGPF for taking care of the financial transactions and accounting. We would also like to thank Sisi Zlatanova for her advice and help in setting up the conference. We appreciate very much the support of Mrs. Agata Oelschla¨ger from Springer Verlag in the planning and preparation of this book. We further thank the scientific organisations ISPRS, EuroSDR, DGPF, and the international standardisation body OGC for taking over patronage of this event. Thanks go also to the program committee and the additional reviewers. We express our gratitude to the sponsors whose contributions helped us much in keeping the registration fees at a reasonable height. Last but not least we thank the invited speakers and all colleagues who have submitted papers about their research and especially those who finally joined the conference and presented their work in either oral or poster presentations. Berlin Germany October 2010

Thomas H. Kolbe Gerhard Ko¨nig Claus Nagel

Organisation

Program Chair Thomas H. Kolbe

Technische Universita¨t Berlin

Scientific Committee Alias Abdul-Rahman Roland Billen Lars Bodum Martin Breunig Eliseo Clementini Volker Coors Ju¨rgen Do¨llner Thomas H. Kolbe Philippe de Maeyer Claudine Me´tral Christopher M. Gold Gerhard Gro¨ger Gerhard Joos Hugo Ledoux Jiyeong Lee Ki-Joune Li Fred Limp Marc-Oliver Lo¨wner Hui Lin Stephan Nebiker ¨ stman Anders O Norbert Pfeifer Jacynthe Pouliot Carl Reed Massimo Rumor Monika Sester Uwe Stilla Andre´ Streilein Rod Thompson

Universiti Teknologi, Malaysia University of Lie`ge, Belgium Aalborg University, Denmark University of Osnabru¨ck, Germany University of L’Aquila, Italy University of Applied Sciences Stuttgart, Germany University of Potsdam, Germany Technische Universita¨t Berlin, Germany Ghent University, Belgium University of Geneva, Switzerland University of Glamorgan, United Kingdom University of Bonn, Germany NATO C3 Agency, The Netherlands Delft University of Technology, The Netherlands University of Seoul, South Korea Pusan National University, South Korea University of Arkansas, Fayetteville, NC, USA Technische Universita¨t Braunschweig Chinese University of Hong Kong, China University of Applied Sciences, Switzerland University of Ga¨vle, Sweden Vienna University of Technology, Austria Universite´ Laval, Que´bec, Canada Open Geospatial Consortium, USA University of Padova, Italy University of Hannover, Germany Technische Universita¨t Mu¨nchen, Germany Swisstopo, Switzerland Queensland Government, Australia

vii

viii Peter van Oosterom George Vosselman Peter Woodsford Mike Worboys Qing Zhu Alexander Zipf Sisi Zlatanova

Organisation Delft University of Technology, The Netherlands ITC Enschede, The Netherlands Snowflake Software, United Kingdom University of Maine, Orono, ME, USA Wuhan University, China University of Heidelberg, Germany Delft University of Technology, The Netherlands

Additional Reviewers Thomas Becker Robert Kaden Claus Nagel

Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany

Local Organizing Committee Thomas Becker Andreas Fuls Javier Herreruela Robert Kaden Gerhard Ko¨nig Andreas Kru¨ger Rosemarie Kunkel Hartmut Lehmann Alexandra Lorenz Claus Nagel Bernd Stary Sven Weisbrich

Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany Technische Universita¨t Berlin, Germany

Contents

Integrated 3D Modeling of Multi-utility Networks and Their Interdependencies for Critical Infrastructure Analysis . . . . . . . . . . . . . . . . . . . . . 1 T. Becker, C. Nagel, and T.H. Kolbe Modeling Space by Stereographic Rejection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 W.L. (Pim) Bil Rapid Modelling of Complex Building Interiors . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 Pawel Boguslawski and Christopher Gold Large Scale Constraint Delaunay Triangulation for Virtual Globe Rendering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 M. Christen and S. Nebiker Towards Interoperating CityGML and IFC Building Models: A Unified Model Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 ¨ stman, and Khurram Shahzad Mohamed El-Mekawy, Anders O Initial Investigations for Modeling Interior Utilities Within 3D Geo Context: Transforming IFC-Interior Utility to CityGML/ UtilityNetworkADE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 Ihab Hijazi, Manfred Ehlers, Sisi Zlatanova, Thomas Becker, and León van Berlo Depth Perception in Virtual Reality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Anja Matatko, Ju¨rgen Bollmann, and Andreas Mu¨ller Interactive Urban and Forest Fire Simulation with Extinguishment Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 ´ lvaro Segura, Anis Korchi, Jorge Posada, Aitor Moreno, A and Oihana Otaegui

ix

x

Contents

3D Cadastre in the Province of Quebec: A First Experiment for the Construction of a Volumetric Representation . . . . . . . . . . . . . . . . . . . . . 149 Jacynthe Pouliot, Tania Roy, Guillaume Fouquet-Asselin, and Joanie Desgroseilliers 3D Modeling for Mobile Augmented Reality in Unprepared Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Vincent Thomas, Sylvie Daniel, and Jacynthe Pouliot Integrated Representation of (Potentially Unbounded) 2D and 3D Spatial Objects for Rigorously Correct Query and Manipulation . . . . . . . 179 Rodney James Thompson and Peter van Oosterom Interactive Rendering Techniques for Highlighting in 3D Geovirtual Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Matthias Trapp, Christian Beesk, Sebastian Pasewaldt, and Ju¨rgen Do¨llner Integration of BIM and GIS: The Development of the CityGML GeoBIM Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Ruben de Laat and León van Berlo Modelling Three-Dimensional Geoscientific Datasets with the Discrete Voronoi Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 Tom van der Putte and Hugo Ledoux Challenges in 3D Geo Information and Participatory Design and Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 Jan B.F. van Erp, Anita H.M. Cremers, and Judith M. Kessens An Integrated Framework for Reconstructing Full 3D Building Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Langyue Wang and Gunho Sohn Towards Semantic 3D City Modeling and Visual Explorations . . . . . . . . . . 275 Qing Zhu, Junqiao Zhao, Zhiqiang Du, Yeting Zhang, Weiping Xu, Xiao Xie, Yulin Ding, Fei Wang, and Tingsong Wang

Contributors

Thomas Becker Institute for Geodesy and Geoinformation Science, Technische Universita¨t Berlin, Berlin, Germany, [email protected] Christian Beesk Hasso-Plattner-Institute, University of Potsdam, Potsdam, Germany, [email protected] W.L. (Pim) Bil Gemeente Amstelveen, Amstelveen, The Netherlands, p.bil@ amstelveen.nl Pawel Boguslawski Department of Computing and Mathematics, University of Glamorgan, Pontypridd, United Kingdom, [email protected] Ju¨rgen Bollmann Cartography, University of Trier, Trier, Germany, [email protected] Martin Christen Institute of Geomatics Engineering, University of Applied Sciences Northwestern Switzerland, Muttenz, Switzerland, [email protected] Anita H.M. Cremers Human Factors, Netherlands Organisation for Applied Scientific Research TNO, Delft, The Netherlands, [email protected] Sylvie Daniel Geomatics Department, Universite´ Laval, Que´bec, QC, Canada G1V 0A6, [email protected] Ruben de Laat Netherlands Organisation for Applied Scientific Research TNO, Delft, The Netherlands, [email protected] Joanie Desgroseilliers Geomatics Department, Universite´ Laval, Que´bec, QC, Canada G1V 0A6, [email protected] Yulin Ding State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] xi

xii

Contributors

Ju¨rgen Do¨llner Hasso-Plattner-Institute, University of Potsdam, Potsdam, Germany, [email protected] Zhiqiang Du State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Manfred Ehlers Institute for Geoinformatics and Remote Sensing, University of Osnabru¨ck, Osnabru¨ck, Germany, [email protected] Mohamed Sobih El-Mekawy Future Position X, Ga¨vle, Sweden, [email protected] Guillaume Fouquet-Asselin Geomatics Department, Universite´ Laval, Que´bec, QC, Canada G1V 0A6, [email protected] Christopher Gold Department of Computing and Mathematics, University of Glamorgan, Pontypridd, United Kingdom; Department of Geoinformatics, Universiti Teknologi, Johor Bahru, Johor, Malaysia, [email protected] Ihab Hijazi Institute for Geoinformatics and Remote Sensing, University of Osnabru¨ck, Osnabru¨ck, Germany, [email protected] Judith M. Kessens Human Factors, Netherlands Organisation for Applied Scientific Research TNO, Delft, The Netherlands, [email protected] Thomas H. Kolbe Institute for Geodesy and Geoinformation Science, Technische Universita¨t Berlin, Berlin, Germany, [email protected] Anis Korchi Vicomtech, Donostia-San Sebastin, Spain, [email protected] Hugo Ledoux GIS Technology Group, Delft University of Technology, Delft, The Netherlands, [email protected] Anja Matatko Cartography, University of Trier, Trier, Germany, [email protected] Aitor Moreno Vicomtech, Donostia-San Sebastin, Spain, [email protected] Andreas Mu¨ller Cartography, University of Trier, Trier, Germany, [email protected] Claus Nagel Institute for Geodesy and Geoinformation Science, Technische Universita¨t Berlin, Berlin, Germany, [email protected]

Contributors

xiii

Stephan Nebiker Institute of Geomatics Engineering, University of Applied Sciences Northwestern Switzerland, Muttenz, Switzerland, [email protected] ¨ stman GIS Institute, University of Ga¨vle, Ga¨vle, Sweden, Anders. Anders O [email protected] Oihana Otaegui Vicomtech, Donostia-San Sebastin, Spain, ootaegui@ vicomtech.org Sebastian Pasewaldt Hasso-Plattner-Institute, University of Potsdam, Potsdam, Germany, [email protected] Jorge Posada Vicomtech, Donostia-San Sebastin, Spain, [email protected] Jacynthe Pouliot Geomatics Department, Universite´ Laval, Que´bec, QC, Canada G1V 0A6, [email protected] Tania Roy Geomatics Department, Universite´ Laval, Que´bec, QC, Canada G1V 0A6, [email protected] ´ lvaro Segura Vicomtech, Donostia-San Sebastin, Spain, [email protected] A Khurram Shahzad Department of Computer and System Sciences, Royal Institute of Technology, Stockholm, Sweden, [email protected] Gunho Sohn GeoICT Laboratory, York University, Toronto, ON, Canada, [email protected] Vincent Thomas Geomatics Department, Universite´ Laval, Que´bec, QC, Canada G1V 0A6, [email protected] Rodney James Thompson Department of Environment and Resource Management, Indooroopilly, Queensland, Australia; OTB, GIS Technology, Delft University of Technology, Delft, The Netherlands, [email protected] Matthias Trapp Hasso-Plattner-Institute, University of Potsdam, Potsdam, Germany, [email protected] León van Berlo Netherlands Organisation for Applied Scientific Research TNO, Delft, The Netherlands, [email protected] Tom van der Putte Netherlands Organisation for Applied Scientific Research TNO, Geological Survey of the Netherlands, Utrecht, The Netherlands, tom. [email protected]

xiv

Contributors

Jan B.F. Van Erp Human Factors, Netherlands Organisation for Applied Scientific Research TNO, Delft, The Netherlands, [email protected] Peter van Oosterom OTB, GIS Technology, Delft University of Technology, Delft, The Netherlands, [email protected] Fei Wang State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Langyue Wang GeoICT Laboratory, York University, Toronto, ON, Canada, [email protected] Tingsong Wang State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Xiao Xie State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Weiping Xu State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Yeting Zhang State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Junqiao Zhao State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Qing Zhu State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China, [email protected] Sisi Zlatanova OTB, Research Institute for Housing, Urban and Mobility Studies, Delft University of Technology, Delft, The Netherlands, [email protected]

.

Integrated 3D Modeling of Multi-utility Networks and Their Interdependencies for Critical Infrastructure Analysis T. Becker, C. Nagel, and T.H. Kolbe

Abstract In today’s technologically advanced society the dependency of every citizen and company on working infrastructures is extremely high. Failures of critical infrastructures, such as the Italian blackout in 2003 or the failure of power supply in wide parts of Europe in 2006, demonstrate the strong linkage of networks across borders. However, also infrastructures within the same geographic region but of different types have strong interdependencies and failures in one type of network can have cascading effects onto the other networks. In order to support risk analysis and planning of emergency response actions the modeling of critical infrastructures and their mutual dependencies in 3D space is required. Decision makers need a comprehensive view of the disaster situation to be able to estimate the consequences of their action. For this purpose, a comprehensive understanding and simulation of cascading or looping effects as well as the propagation of the disaster extend is needed. But neither the existing utility networks models nor the international standards for modeling cities or buildings map the mutual interrelationships between different infrastructures or between the city and its infrastructures. In this paper the requirements and a novel framework for the integrated 3D modeling of critical infrastructures within cities is presented. By giving a dual representation utility network components are modeled both according to their 3D topography and by a complementary graph structure embedded into 3D space.

1 Introduction The field of critical infrastructures, the simulation of their failure and breakdown, and possible occurrences of cascading or looping effects are examined by researchers worldwide.

T. Becker (*), C. Nagel, and T.H. Kolbe Institute for Geodesy and Geoinformation Science, Technische Universit€at Berlin, Strasse des 17, Juni, 10623 Berlin, Germany e-mail: [email protected]; [email protected]; [email protected]

T.H. Kolbe et al. (eds.), Advances in 3D Geo-Information Sciences, Lecture Notes in Geoinformation and Cartography, DOI 10.1007/978-3-642-12670-3_1, # Springer-Verlag Berlin Heidelberg 2011

1

2

T. Becker et al.

The European Union, Germany, United Kingdom, and the USA have criticalinfrastructure programmes which mostly define electricity generation, transmission, and distribution as well as gas production, transport and distribution, telecommunication, water supply, transportation and heating, and many more as critical infrastructures (see Pederson et al. 2006). These programmes address the identification of important objects, risk analysis based on threat scenarios (Johnson 2007; Dudenhoefer et al. 2006) and the vulnerability of each object as well as the identification, selection, and priorisation of counter-measures and procedures. As it is shown by Johnson (2007), a small fault can trigger a blackout of a whole power supply network across the borders of different countries, leading to a domino effect that has separated, in the discussed case Italy from the whole European grid. It is easily conceivable that many other infrastructures are critically depending on electrical energy. A wide failure of energy supply would have for example the consequence that telecommunications, water supply, sewage disposal and many other infrastructures would also fail, assuming that they do not dispose of a backup system. However, it is completely unknown whether, e.g., by the failure of water supply (which may include uncontrolled leakage of water) failures in other infrastructures can be induced (cascade effect) or even are amplified and reflected (loop effect), or which effects or damages will be caused on the city structures. In order to be able to estimate the consequences of their action, decision makers have to get a comprehensive view of the disaster situation. A very detailed understanding and simulation of cascading or looping effects as well as the propagation of the disaster extend is needed. The base for such a common simulation and an all comprehensive common operational picture (COP) of the actual situation of a disaster is an integrated 3D topographic model of critical infrastructures and their inherent spatial relations to the city. In Sect. 2 the requirements on modeling 3D multi-utility networks as well as modeling of interdependencies between critical infrastructures are discussed in detail. Hence, in Sect. 3, an overview of related models like the Industry Foundation Classes (IFC) (Liebich 2009), ArcGIS and INSPIRE is given, whereas in Sect. 4 our conceptual approach of modeling utilities with respect to the identified requirements is presented. In Sect. 5 it is shown how the UtilityNetworkADE is implemented as an extension of the international standard CityGML (Gr€oger et al. 2008). In Sect. 6 we draw conclusions and give an outlook to future work.

2 Requirements on Modeling 3D Multi-utility Networks Pederson et al. (2006) and Tolone et al. (2004) present some scenarios and approaches to model and simulate critical infrastructures and their complex interdependencies. Tolone et al. further demonstrate that the behavior of critical infrastructures can be better understood through multi-infrastructure simulations. In order to understand the dynamics and dependencies of utility networks and to allow the propagation and visualization of effects (see Johnson 2007) a first identified requirement

Integrated 3D Modeling of Multi-utility Networks and Their Interdependencies

3

on multi-utility networks is that a geometrical, topological, and functional embedding of critical infrastructures into the urban space must be done. This will also allow the joint visualization of virtual 3D city models and 3D utility networks, which would be very helpful to understand the locations and spatial relations of infrastructures in the context of city objects. A better understanding facilitated by a joint 3D visualization would result in a better decision making process and coordination between decision makers and actors such as fire fighters. Moreover, an integrated modeling of utilities and city topography facilitates the visualization and analysis of dependencies between city objects and network components, such as which city object depends on which infrastructure. Complex analyses or simulations such as collision detection (e.g. excavator vs. pipe), determination of explosion impact (determination of damaged objects), and simulations predicting, for example, the spread of water in a flood scenario above and below the ground require the 3D topographic representation and description of the components of the utility network of a city. The structural description of the entire network by graph structures with a 3D embedding is suitable to calculate the relative position between network objects of different commodities, their spatial relation to each other as well as their spatial relation to other objects of the city, such as buildings, city furniture, or roads. Due to the fact that the different types of infrastructure of the city lie above and in between each other the embedding into the 3D space plays an important role. Furthermore, 3D visual inspection helps in getting a better understanding of the spatial relations of the networks relative to each other. Network analyses such as the calculation of slope or slope change becomes possible. Thus, it is conceivable that a 3-dimensional description of the city (see Gr€ oger et al. 2008) as well as a suitable 3D description of the underlying utility network has to be realized. The 3D objects of the network must be integrated into the 3D space of the city and thus they can be queried in the context of a disaster management. An investigation of existing utility network models (ESRI 2003, 2007; INSPIRE Data Specifications Drafting Team 2009; INSPIRE Thematic Working Group Transport Networks 2009; Meehan 2007; Liebich 2009; Becker et al. 2010) shows that different types of hierarchies are used to decompose networks into sub networks or single features into their components. On the one hand hierarchies are suitable to express different kind of pressure levels, flow rates as well as voltage levels. On the other hand hierarchies are suitable for realizing different kinds of feature aggregation within the same sub network. While for the evaluation of the connectivity of network objects a very general view of the network connectivity is sufficient, simulating the failure of network components and resulting cascading effects may require a more detailed description of the interior of a network component. For example, a switch gear cabinet may consist of different switches, transformers, and fuses. Each of these components of the switch gear cabinet may trigger a network failure and may lead to a cascading effect across different networks. Thus, it is mandatory for an integrated network model that both the whole network can be expressed as an aggregation of different sub networks with homogeneous commodity and those individual network components, such as

4

T. Becker et al.

a pump or switch gear cabinet can be represented as an aggregation of a set of network components. In the context of disaster management and the analysis of critical infrastructures it is important to understand that critical infrastructures interact with each other, either by direct connection, or due to effects resulting from their spatial proximity. These interactions are called interdependencies and can be classified in different types. Some type definitions for interdependencies are given by Dudenhoefer et al. (2006). They categorize interdependencies as physical – direct linkage, geospatial – infrastructure components at the same location (only 2D!), policy – binding of infrastructure components due to policy or high level decisions as well as, informational – a binding of information flow between infrastructures. The interactions between infrastructures are often very complex and may cause cascading effects which can lead to disasters. Therefore the modeling and development of models that accurately simulate critical infrastructure behavior and their interdependencies and vulnerabilities is important (see Laprie et al. 2007; Klein 2009; Klein et al. 2009; Setola and Geretshuber 2008; Liebich 2009). Other research groups have developed various modeling approaches including agent based modeling (Usov and Beyel 2008; Klein 2009; Klein et al. 2009), effects-based operations modeling (Johnson 2007; Dudenhoefer et al. 2006), mathematical models and models based on risk (Pederson et al. 2006; Setola and Geretshuber 2008). But as mentioned before, in chaotic situations like disasters or emergencies, decision makers should understand the dynamics and dependencies of infrastructures. Failure to understand those dynamics will result in ineffective response, domino effects (see Johnson 2007) and poor coordination between decision makers and emergency actors. Thus, Dudenhoefer et al. (2006) present in their paper the issues of interdependency modeling, simulation, and analysis. They examine the infrastructure interdependency and provide additionally some formalism for these dependencies. Even though it is the most comprehensive work concerning interdependencies and critical infrastructures (to the best of our knowledge), the linkage to objects and topography of the city and its resulting effects to the city are neglected. As a result of the previous discussion an integrated 3D model of heterogeneous utility networks has to meet the following requirements: 1. Utility networks must be embedded into 3D space and need to be integrated with 3D representations of urban entities, i.e. 3D city models. 2. Network components must be represented both by 3D topographic descriptions and by a topological and structural description of all network components with an embedding in 3D space. 3. In general, the model must support both types of hierarchical modeling: the modeling of feature hierarchies (as an aggregation of many features) and the modeling of network hierarchies (as an aggregation of sub networks of the same type of commodity). 4. An integrated model for multi-utility networks must support the simultaneous representation of heterogeneous networks.


5

5. Special attention has to be paid to the modeling of interdependencies, which establish explicit relations between the network entities of different utility types and can be of spatial, non-spatial, logical, informal, and policy kind (cf. Dudenhoefer et al. 2006). These requirements form the basis for the examination of existing concepts, systems, and the development of a new comprehensive framework in the following sections.

3 Related Models Several models for utility networks on the urban and building scale have been developed in the past. In the following, we will examine the utility models of the Industry Foundation Classes (IFC), the commercial GIS ArcGIS, and the recently presented network core model of the European INSPIRE initiative.

3.1

Modeling Utilities in IFC

The most important standard for data exchange of buildings in the field of architecture and civil engineering are the Industry Foundation Classes (IFC) (Grise et al. 2001). The IFC represent logical building structures, accompanying properties, optional 2D and 3D geometry as well as utilities. The module “Shared building service elements” is defined in the interoperability layer and defines the basic concepts for interoperability between building service domain extensions, such as electrical and control domains or heating, ventilation, air conditioning, etc. This module includes basic type definitions as subtype of IfcTypeProduct, for instance sanitary types, boiler types, etc. and occurrence definitions for flow and distribution systems which specify the placement (position) data, instance specific data, references the type and finally the connectivity with other elements and systems. In order to build up a network the IFC offer two different ways of connectivity. A connection between building service elements may be physical or logical. In general, a logical connection realizes the connection of two components via Ports, whereas a physical connection realizes the connection via a realizing element such as IfcFlowFitting. The connectivity concept of IFC comprises both the physical connection between elements (IfcRelConnectsElements) and the logical connection of building service items on the level of their ports (IfcRelConnectsPorts). An IfcPort can have a placement and a representation and occurs at each connection point between IfcElements. The connectivity is realized through the use of IfcRelConnectsPorts and only compatible elements are allowed to connect to each other. If the optional attribute RealizingElement of IfcRelConnectsPorts is used the connectivity is switched from a logical connectivity to a physical connectivity

6

T. Becker et al.

Fig. 1 Connectivity example between two building service elements (Liebich 2009)

(see Fig. 1). In order to define the flow direction it is possible to use the IfcFlowDirectionEnum, which defines a connection point as a “source of flow”, a “flow sink”, as a “source and sink”, or as “undefined”. Element hierarchies can be expressed by using the IfcRelDecomposes relationship which supports the concept of elements being composed or decomposed. This realizes a whole/part hierarchy which allows traversing from the whole to the parts and vice-versa. Thus, a pump may be the aggregation of IfcFlowFitting, IfcFlowController, and IfcFlowMovingDevice. While it is possible to express feature hierarchies, network hierarchies cannot be represented. This may be due to the fact that IFC is focused on the representation of single buildings (or building complexes). However, it is possible to model the interdependencies between different networks by using the logical connection of IfcRelConnectsPorts (see Liebich 2009, p. 88). In summary, IFC supports the representation of the 3D topography of network elements and the logical connectivity on the level of ports within a graph structure. However, there is no explicit embedding of the graph in 3D space; the location is only given implicitly, if network components have a 3D geometry. The utility network model focuses on structures within sites and does not address networks on the city level interconnecting sites/buildings.

3.2

Modeling Utilities in ArcGIS

In ArcGIS the different types of utility networks are mapped to a core model which is based on geometric networks, being labeled graphs with a spatial embedding. This core model represents the basic structure for all kinds of utilities. A network is constructed from edges and junctions as line and point features. Each real world utility object can be represented as one feature in the network whereas same kinds of


7

Main line

features can be represented by a feature class (Bedford 2004; ESRI 2003, 2007; Grise et al. 2001). The geometric part of a utility network (see Fig. 2) is a single graph structure consisting of Edge and Junction elements and can have any number of participating feature classes. The graph is composed of features from one or more feature class in a feature data set (ESRI 2003). It binds together feature classes which form a network and contains all attributes, relationships, and validation rules. The logical network (see Fig. 2) is a special data structure to store the connectivity between features of the network and is implemented by a set of tables. This network is automatically set up and maintained when the geometric network is created or edited. In contrast to ESRI’s street network model all feature classes can take part in the networks as one of the four network types: simple edge, simple junction, complex edge, and complex junction. A simple edge has a one-to-one relation between the feature and the edge element and connects always two junctions. However, a complex edge may connect more than two junctions in order to represent a sequence of edge elements divided by junctions. While they realize a logically connected sequence of

Meter vault Service lateral

Pump

Main line Valve

Tap

Hydrant Geometric network

Valve 2

Valve Tee

Valve 3

Pump Check Valve 1

Hydrant

logical network

Fig. 2 Example of a feature dataset consisting of a geometric and a logical network

8

T. Becker et al.

edges, they geometrically represent a single feature – and in this respect it is a kind of feature hierarchy. Simple junctions have a one-to-one correspondence between the feature and the junction element and are represented only by one point in the network. In order to enhance the feature hierarchy approach a complex junction can be represented as an object in the geometric network and as multiple objects including junctions and edges in the logical network. A pump for example may be represented as a box in the geometric network with one input line and two output lines. However in the logical network this box is represented as a set of valves, tees, pumps (represented by nodes), and pipes (represented by edges); see lower part of Fig. 2. In order to support the flow of commodity a junction feature can have the attribute AncillaryRole, representing its role as source, sink, or neither. To prevent that non-suitable network components are linked to each other, connectivity rules can be specified. The modeling of network hierarchies as well as the modeling of interdependent utility networks within the framework (as it is possible in the transportation network model of ArcGIS which supports multi-modal networks) is not supported yet.

3.3

The Generic Network Model of INSPIRE

The “Network” package of the INSPIRE application schemas (INSPIRE Data Specifications Drafting Team 2009) defines the basic application schema for networks which is extended by additional, domain specific spatial data schemas (INSPIRE Thematic Working Group Transport Networks 2009). The central class is NetworkElement which may be any entity that is relevant for a network. The network package consists of further classes, which are required for modeling networks, such as Network, Link, and Node. A Network is a collection of NetworkElements which is the superclass for elements like Area, Node, and some special classes such as GeneralisedLink, LinkSet and GradeSeparatedCrossing (see Fig. 3). Thus, a simple network may only consist of Nodes and Links, where a Link must be bounded by exactly two nodes. The clear distinction of NetworkElements into point like (Node) and line like (Link) objects and the lack of a feature aggregation schema does not allow for hierarchical decompositions of network components. For example, a point like object cannot consist of other point like or line like objects. The hierarchical modeling of a line like object is supported by the class LinkSet. A hierarchical modeling of a network and the modeling of interdependencies is possible by using the class NetworkConnection. NetworkConnections are also NetworkElements relating two or more arbitrary NetworkElements facilitating the modeling also of hierarchical networks (see INSPIRE Thematic Working Group Transport Networks 2009, p. 93). Since the INSPIRE data model specifications do not make use of 3D geometries, the graph can only be embedded in 2D. Also the INSPIRE core network model does not consider a dual representation of network entities by a 3D topographic and 3D network model.


9

Fig. 3 Network Application schema of INSPIRE (Adapted from INSPIRE Data Specifications Drafting Team 2009)

4 3D Multi-utility Network Model In this chapter we propose a novel framework for the representation of 3D multi-utility networks which overcomes the limitations of existing modeling approaches discussed in the previous chapter and addresses the requirements for critical infrastructure models as identified in Chap. 2. The entire data model will be given in UML in Chap. 5.

4.1

Conceptual Modeling

We define a utility network as abstraction of the collection of all topographic realworld objects being relevant network components such as pipes, pumps, or switchgear cabinets. The conceptual modeling of a utility network and its components employs the ISO 19100 series of geographic information standards for the modeling of geographic features issued by ISO/TC 211. According to the General Feature Model (GFM) specified in ISO 19109 (ISO/TC 211 2008), geographic features are defined as abstractions of real world objects. The GFM is a metamodel that

10

T. Becker et al.

introduces general concepts for the classification of geographic features and their relations as well as the modeling of spatial and non-spatial attributes and operations. Object oriented modeling principles can be applied in order to create specialization and aggregation hierarchies.

4.1.1

NetworkFeature and FeatureGraph

The basic unit for modeling utility networks within the proposed framework is NetworkFeature. NetworkFeature is an abstract concept which maps topographic network components onto respective GFM feature types. It allows for the modeling of thematic properties as well as the definition of taxonomies and partonomies of network components. Hence, it forms the base for the semantic modeling of concrete network features in various infrastructures such as gas, power, or water supply networks. A main concept of our modeling approach is the dual representation of a NetworkFeature according to which each network component can be represented both by its 3D topography and by means of a complementary graph structure called FeatureGraph. The dual representation addresses the need for both types of representation as discussed in Chap. 2. The modeling of 3D topography is realized in compliance with ISO 19109 as spatial aspect of a NetworkFeature. The value domain for spatial attributes is defined by the Spatial Schema specified in ISO 19107 (Herring 2001) which allows for describing the geometry of a NetworkFeature in up to three dimensions. In addition to its 3D geospatial representation, each network component is mapped by FeatureGraph onto a separate graph structure representing its functional, structural, and topological aspects. This concept differs substantially from previous modeling approaches which usually map the entire utility network onto a single topological network graph derived from a classification of network components into point-like and line-like objects. For example, pipes within a water supply infrastructure are usually given as line-like features and thus mapped onto edges. A T-fitting element is consequently to be mapped onto a node at the intersection of its connected pipes. Functional or structural aspects of the T-fitting are not taken into account. For example, its pipe-connecting ends could also be represented as individual edges, especially if they have a considerable geometric extent or in order to model structural changes such as differing pipe diameters. However, existing approaches lack the possibility to map a single network component onto a set of nodes and edges. In contrast, the proposed FeatureGraph explicitly allows for a graph-based representation of each NetworkFeature. The resulting graph structure follows the general principles of graph theory (Diestel 2005) and has clear semantics. We distinguish two types of nodes: exterior and interior nodes. An exterior node represents a topological connection point where other network components may be attached. Interior nodes are used to model internal structural, physical, or logical aspects of the network component. For example, an interior node may represent the


11

narrowing of a water pipe which results in changes of flow speed and pressure. In contrast to exterior nodes, no other network component may be topologically connected to an interior node. In order to link a pair of nodes within the graph structure of a NetworkFeature, a special type of edge called InteriorFeatureLink is introduced which is only allowed to relate two nodes belonging to the same FeatureGraph instance. Figure 4 exemplifies two valid FeatureGraph representations for a T-fitting element. The first alternative results from a functional description of the T-fitting and represents its pipe-connecting ends as individual edges. Each connection point is explicitly marked by an exterior node. An interior node is added where the two axis of the T-fitting meet. The second alternative maps the entire T-fitting to a single exterior node which is the minimum possible FeatureGraph representation. Modeling a single InteriorFeatureLink is not allowed because an edge has always to be bounded by exactly two nodes. In order to support sophisticated graph-based analyses and simulations, both nodes and edges may carry additional attributes, e.g. to model weights or the FlowControl of a NetworkFeature. Examples for FlowControl are the on/off state of a switch within an electrical circuit or the gradual impact of a valve on the flow of water. Furthermore, the exterior nodes of the FeatureGraph may denote a ConnectionSignature. Only those network components having an identical or compatible ConnectionSignature are allowed to connect. A ConnectionSignature is to be seen as set of connectivity constraints addressing arbitrary aspects of a NetworkFeature such as functional, physical, logical, or even geometric properties. For example, the exterior nodes of a T-fitting could define a certain pipe diameter as

Alternative A:

Legend FeatureGraph

Node (type: exterior) Node (type: interior)

Alternative B:

InteriorFeatureLink NetworkFeature

FeatureGraph

Fig. 4 Two alternative FeatureGraph representations for the same object

12

T. Becker et al.

required connection signature for connecting elements. Both FlowControl and ConnectionSignature are abstract concepts which have to be specified within the context of a concrete utility network model. Additionally, each FeatureGraph can have a geometric embedding in 3D space which allows for 3D analyses of the FeatureGraph instances and introduces metric information such as edge lengths into the graph. Due to the semantic differentiation of nodes and edges and the possibility to model attributes, the resulting graph can be classified as both typed and attributed. This requires the conceptual modeling of nodes and edges as semantic objects. Consequently, the topological primitives TP_Node and TP_Edge specified by ISO 19107 are infeasible for their representation. Nodes and edges are rather mapped onto two corresponding GFM feature types called Node and Edge. By this means, nodes and edges can be semantically classified and augmented by spatial and nonspatial attributes. The geometric embedding of both Node and Edge is realized as spatial realization association to a GM_Point respectively a GM_Curve primitive. A FeatureGraph is modeled as feature collection of Node and Edge instances.

4.1.2

Network and NetworkGraph

The aggregation of NetworkFeature instances to an entire utility network is called Network. A Network is characterized by a homogeneous type of commodity such as water, electricity, or gas. Analogously to NetworkFeature, a Network employs the concept of dual representation. Whereas its 3D topography is implicitly given by the 3D topography of its components, the graph-based mapping is explicitly modeled as NetworkGraph. A NetworkGraph represents the graph structure of the entire utility network. For this purpose, it links the FeatureGraph instances of the aggregated network features. Thus, a NetworkGraph conceptually is to be seen as graph of graphs. It results from the pair-wise linking of exterior nodes in different FeatureGraph instances being parts of the same Network. For this purpose, a special subtype of Edge is introduced called InterFeatureLink. The following figure depicts two valid excerpts of NetworkGraph representations. For illustration purposes, the connected pipe elements are shown in a detached way. The main purpose of InterFeatureLink is to establish a topological connection between two network features. Since it is derived from Edge, it can additionally be embedded into 3D space. However, a further spatial representation of an InterFeatureLink is not required if both exterior nodes linked by the InterFeatureLink geometrically collapse (see the first alternative of Fig. 5). Network and NetworkGraph are the central concepts within our framework allowing for analyses and simulations of an entire utility network based on its 3D topography and graph-based representation. Both are realized as GFM feature collections.


13

Alternative A:

FeatureGraph

FeatureGraph NetworkGraph

Alternative B:

FeatureGraph

Legend Node (type: exterior) Node (type: interior) InteriorFeatureLink InterFeatureLink NetworkFeature

FeatureGraph

NetworkGraph

Fig. 5 Two different modeling alternatives for a NetworkGraph connecting two FeatureGraphs using an InterFeatureLink

4.1.3

Modeling Hierarchies

Our framework supports hierarchical modeling for both network components and entire utility networks. A component hierarchy allows for the decomposition of a NetworkFeature into its parts which again are components of the utility network. Thus, it represents a consists-of-relation between two or more NetworkFeature instances of the same Network and is conceptually realized through an aggregation association of NetworkFeature with itself. This recursive modeling approach enables hierarchies of arbitrary depth. For example, a switchgear cabinet is a NetworkFeature of a power supply network which provides connectors to attach other network components. At the same time, it is internally built from further devices such as switches controlling the flow of electricity. These internal components are both connected to the switchgear cabinet and to each other, and may themselves consist of further internal devices. Figure 6 sketches this example. For its graph-based representation, a feature hierarchy is simply a set of FeatureGraph instances which have to be topologically connected at their exterior nodes. Again, this is to be seen as a graph of graphs. For linking FeatureGraph instances, the InterFeatureLink has been introduced in the previous section. However, the concept of the InterFeatureLink has to be augmented such that it additionally provides information about whether the network components are connected on the same level or over different levels of the hierarchy. Graph traversal algorithms

14

T. Becker et al.

Switch gear cabinet Legend A1

A

A 1

B5

3

B3 A2 1

5

6

Node (type: exterior) Node (type: interior)

C1

2 4 D4

4

B4 B1

C4 3

2

C3 5 D3

InteriorFeatureLink InterFeatureLink (connects) B2

6 D2 D1

C2

InterFeatureLink (contains) NetworkFeature

Fig. 6 Feature aggregation and use of InterFeatureLinks

can evaluate this information in order to decide whether or not to traverse down a hierarchy. The modeling of hierarchies of utility networks follows the same concepts. A network hierarchy is used to semantically and topologically decompose a Network into sub networks all of which share a homogeneous type of commodity. For example, a gas supply network can be separated into three sub networks for high pressure, medium pressure, and low pressure. However, in order to establish a link between two FeatureGraph representations within different parts of the hierarchy, an InterFeatureLink is infeasible since it may only connect network features within the same Network (cf. previous section). For this reason, a third subtype of Edge called NetworkLink is defined which has to be used to connect the exterior nodes of two FeatureGraph instances when crossing Network borders.

4.1.4

Multi-utility Networks and Interdependencies

In contrast to the existing modeling approaches introduced in Chap. 3, the proposed model explicitly facilitates the integration of multiple utility networks with heterogeneous type of commodity. Each utility network is modeled as separate instance of Network following the general concepts introduced in the previous sections. As shown in Chap. 2, interdependencies between infrastructures may result from various reasons such as physical linkage, geospatial adjacency or closeness, or policy. The different types of interdependencies can be introduced into our model through the concept of dual representation and, thus, are available for cross-network simulations and analyses. First, geospatial effects can be implicitly evaluated based on the 3D topography representations of both network features and networks. Second, interdependencies can be made explicit within the graph-based


15

representation of the multi-utility network as additional link between FeatureGraph instances belonging to different utility networks. Conceptually, this link is modeled as NetworkLink. Since NetworkLink is realized as GFM feature type, it may carry arbitrary further spatial and non-spatial properties. By this means, a NetworkLink can be augmented by additional information in order to model a specific type of interdependency such as a physical, informational, or policybased interdependency.

5 Mapping to a CityGML Application Domain Extension The integration of utility network models into their 3D urban context in order to capture their mutual impact on urban objects has been identified a crucial requirement for simulations and emergency studies in the field of disaster management (cf. Chap. 2). Semantic 3D city models provide the corresponding integration platform. CityGML (Gr€oger et al. 2008) is an international standard for the representation and exchange of virtual 3D city and landscape models issued by the Open Geospatial Consortium (OGC). CityGML defines an Urban Information Model which represents the semantics and ontological structure of the most relevant topographic objects in cities and regional models in addition to their 3D geometry, topology and appearance information. The conceptual model of CityGML is based on the ISO 19100 standards family and is implemented as an application schema for OGC’s Geography Markup Language (GML 3.1.1) (Cox et al. 2004). CityGML has been designed as a universal topographic information model that defines the feature classes and attributes which are useful for a broad range of applications. It provides thematic models for the representation of buildings, digital terrain models, city furniture, land use, water body, and transportation, etc. For extra information to be modeled and exchanged which is not covered by the thematic models CityGML offers an extension mechanism called Application Domain Extensions (ADE). An ADE specifies a systematic extension of the CityGML data model comprising the introduction of new properties to existing CityGML classes as well as the definition of new feature types or entire conceptual models based on the general concepts of CityGML. In order to embed the 3D multi-utility network model into the context of a 3D city model and to enrich CityGML by the possibility to represent utility networks, we have mapped our conceptual model onto a CityGML ADE called UtilityNetworkADE. Figure 7 illustrates the structure of the ADE by means of a UML package diagram. The proposed conceptual model for 3D multi-utility networks is contained in the core package NetworkCore. It has a dependency relation to CityGML and to the GML3 geometry model which implements the ISO 19107 standard and is used for modeling the 3D topography of network features. Based on the abstract concepts of the NetworkCore package, further concrete packages for different types of utility networks can be implemented such as GasNetwork or FreshWaterNetwork.

16

T. Becker et al.

CityGML::Core

GML 3.1.1

Utility NetworkADE NetworkCore +_NetworkFeature + FeatureGraph + Network + NetworkGraph ...

NetworkComponents + Pipe + Connector ...

SweepGeometry

GasNetwork

ElectricalPowerNetwork

FreshWaterNetwork

WasteWaterNetwork

Fig. 7 Package diagram of UtilityNetworkADE

Common types of network features which can be reused within these packages are grouped into NetworkComponents. Future steps will include the specification of these packages. Finally, Fig. 8 shows the 3D multi-utility network model as UML class diagram. The class names follow the names of the respective concept they implement as introduced in Chap. 4. Prefixes are used to denote both CityGML and GML classes whereas the stereotypes and indicate the corresponding general concepts from ISO 19109 and 19107. The abstract base class _NetworkFeature for modeling network components is derived from the thematic CityGML root class core::_CityObject. First, by these means the UtilityNetworkADE is embedded into the CityGML data model. Second, the general modeling concepts of CityGML for the representation of topographic objects also apply to _NetworkFeature. An entire utility network is represented by the class Network as an aggregation of _NetworkFeature objects and is realized as gml::_FeatureCollection. The proposed UtilityNetworkADE has already been successfully employed by Hijazi et al. (2010) for mapping interior utility structures modeled in IFC to our conceptual framework without information loss. For this purpose concrete network feature classes based on the NetworkCore package have been defined.

Integrated 3D Modeling of Multi-utility Networks and Their Interdependencies core::_CityObject

gml::_Feature

17

> NodeType +exterior +interior

gml::_FeatureCollection

> InterFeatureLinkType _NetworkFeature

0..*

Network

* component

+targetCityObject : anyURI [0..1]

1

1 1

1

0..* topoGraph

0..1 0..*

FeatureGraph 1

0..* topoGraph

subNetwork

consistsOf

1

1

nodeMember realization

1

0..1

start 0..*

realization 0..11

1

end

1

0..*

_Edge + direction : gml::SignType [0..1] +linkControl : _FlowControl [0..1]

linkMember

0..*

0..* 0..*

InteriorFeatureLink

linkMember

Constraints: Both nodes must belong to the same FeatureGraph

_FlowControl

+type : NodeType +connectionSignature : _Signature [0..1] +linkControl : _FlowControl [0..1]

gml::Point

gml::_Curve

0..1

1

Node

1..*

_Signature

1

NetworkGraph

featureGraphMember

+connects +contains

InterFeatureLink

linkMember

NetworkLink

+type : InterFeatureLinkType

Constraints: 1. Each node must belong to a different FeatureGraph 2. Each node type must be exterior 3. The connectionSignature of both nodes must be compatible / identical 4. Both nodes must belong to the same Network

Constraints: 1. Each node must belong to a different FeatureGraph 2. Each node type must be exterior 3. The connectionSignature of both nodes must be compatible / identical 4. Each node must belong to a different Network

Fig. 8 UML class diagram of the UtilityNeworkADE core model

6 Conclusions and Outlook Comprehensive analyses of critical infrastructure require the integrated and simultaneous consideration of multiple heterogeneous utility networks within a common framework. Only by these means mutual dependencies between infrastructures can be evaluated that may cause cascading effects across different networks. A key prerequisite for corresponding network models is the possibility to link critical infrastructures in order to denote interdependencies explicitly. Furthermore, the model must support geospatial analyses in order to determine the implicit interdependencies based on spatial relations like proximity between network components within the same or different infrastructures. For example, the burst of a water

18

T. Becker et al.

pipe may cause damage to nearby or underlying components of a power supply network. Such geospatial analyses require the embedding of network structures into 3D space as well as the 3D topographic representation of network components in addition to a graph-based model. Moreover, the topographic representation allows for the integration of network components into their 3D urban context in order to capture the mutual influence of urban objects. None of the existing models for utility networks examined in Chap. 3 supports all of these aspects. For this reason, in Chap. 4 of this paper we have presented a novel framework for the integrated 3D modeling of critical infrastructures which meets the requirements. Its fundamental concept of dual representation facilities the representation of a utility network component both by its 3D topography and by a complementary graph structure which can be embedded into 3D space. In contrast to other approaches, this graph structure is not derived from a classification of network components into either point-like or line-like objects. We rather employ a stronger concept which allows the mapping of each component onto its own separate graph structure in order to be able to model further structural, physical, or logical aspects. The graph representation of the entire utility network results from topologically connecting these componentbased graphs. The modeling of hierarchies of arbitrary depth is supported on the level of both single network objects and entire utility networks. Mutual dependencies between infrastructures can be introduced as further edges between the graph representations of heterogeneous networks. Finally, we have mapped the proposed conceptual model onto a CityGML Application Domain Extension called UtilityNetworkADE in order to integrate the utility networks with 3D city models (cf. Chap. 5). The proposed framework forms a common core for multi-utility simulations and analysis of interdependent critical infrastructures. As it is a superset of the utility network models examined in Chap. 3, datasets represented according to these models can be converted into the new framework without information loss. As for future work, we will show how existing datasets modeled according to these approaches can be mapped onto our framework. For this purpose, we will augment the developed core model of the UtilityNetworkADE with packages for concrete network objects of the different types of utilities such as water and gas pipes, cables, valves, pumps, and switchgear cabinets. Moreover, we will further investigate different types of network interdependencies, e.g. policy-based or informational, as well as their modeling requirements in order to map them onto the abstract concepts of our framework. It is planned to bring the UtilityNetworkADE into the future revision process of CityGML in order to make it an official module of this international standard. Acknowledgements The presented work was mainly carried out within the collaboration project “SIMKAS 3D” funded by the Federal Ministry of Education and Research of Germany. Additionally, we would like to thank the modeling group of the Special Interest Group 3D of the German National Spatial Data Infrastructure (GDI-DE) for the cooperation and fruitful discussions. We also thank Hartmut Lehmann for his help with the illustrations.


19

References T. Becker, C. Nagel, T. H. Kolbe (2010): UtilityNetworkADE – Core Model. Draft version. Online available at http://www.citygmlwiki.org/index.php/CityGML_UtilityNetworkADE, last access 7. 5. 2010. M. Bedford (2004): GIS for water management in Europe. ESRI Press, Redlands, CA. S. Cox, P. Daisy, R. Lake, C. Portele, A. Whiteside (2004): OpenGIS Geography Markup Language (GML) Implementation Specification V3.1.0, OGC Doc. No. 03-105r1. R. Diestel (2005): Graph theory. 3rd edition, Series on Graduate Texts in Mathematics 173, Springer, Berlin. D. Dudenhoefer, M. Permann, M. Manic (2006): CIMS: A Framework for Infrastructure Interdependency Modeling and Analysis. In: L. F. Perrone, F. P. Wieland, J. Liu, B. G. Lawson, D. M. Nicol, R. M. Fujimoto (eds.), Proceedings of the 38th Conference on Winter Simulation, Monterey, CA. ESRI (2003): ArcGIS Water Utility Data Model. Published by Environmental Systems Research Institute, Redlands, CA. Online available at http://www.downloads2.esri.com/resources/ datamodels/ArcGISWaterUtilityDataModel.pdf, last access 13. 4. 2010. ESRI (2007): GIS Technology for Water, Wastewater, and Storm Water Utilities. Published by Environmental Systems Research Institute. Online available at www.esri.com/library/ brochures/pdfs/water-wastewater.pdf. S. Grise, E. Idolyantes, E. Brinton, B. Booth, M. Zeiler (2001): Water Utilities. ArcGIS™ Data Models. Environmental Systems Research Institute. http://www.downloads2.esri.com/ resources/datamodels/ArcGIS_Water_Utilities.zip, last access 13. 4. 2010. G. Gr€oger, T. H. Kolbe, A. Czerwinski, C. Nagel (2008): OpenGIS City Geography Markup Language (CityGML) Encoding Standard. Version: 1.0.0, OGC Doc. No. 08-007r1, Open Geospatial Consortium. J. Herring (2001): The OpenGIS Abstract Specification. Topic 1: Feature Geometry (ISO 19107 Spatial Schema). Version 5. OGC Doc. No. 01-101. I. Hijazi, M. Ehlers, S. Zlatanova, T. Becker, L. van Berlo (2010): Initial Investigations for Modeling Interior Utilities Within 3D Geo Context: Transforming IFC Interior Utility to CityGML UtilityNetworkADE. In: T. H. Kolbe, G. K€onig, C. Nagel (eds.), Advances in 3D GeoInformation Science, LNG&C Series, Springer, Berlin (this book). INSPIRE Data Specifications Drafting Team (2009): INSPIRE Generic Conceptual Model (D2.5: Generic Conceptual Model, V3.2). Online available at http://www.inspire.jrc.ec.europa.eu/ documents/Data_Specifications/D2.5_v3.2.pdf, last access 13.04.2010. INSPIRE Thematic Working Group Transport Networks (2009): D2.8.I.7 INSPIRE Data Specification on Transport Networks – Guidelines. Online http://www.inspire.jrc.ec.europa.eu/documents/ Data_Specifications/INSPIRE_DataSpecification_TN_v3.0.pdf, last access 13.04.2010. ISO/TC 211 (2008): Geographic Information – Rules for application schema. ISO 19109:2005. International Organization for Standardization (ISO). C. W. Johnson (2007): Analysing the Causes of the Italian and Swiss Blackout, 28th September 2003. In: T. Cant (ed.), Proceedings of the 12th Australian Conference on Safety-Related Programmable Systems (SCS 2007), Adelaide, Australia (30–31 August 2007), Conferences Research and Practice in Information Technology (CRPIT) Vol. 86, pp 21–30. R. Klein (2009): Information Modelling and Simulation in Large Dependent Critical Infrastructures. An Overview on the European Integrated Project IRRIIS. In: Proceedings of the 3rd International Workshop on Critical Information Infrastructures Security, CRITIS 2008, Rome, Italy, October 13–15, 2008, LNCS 5508, Springer, Berlin. R. Klein, E. Rome, C. Beyel, R. Linnemann, W. Reinhardt, A. Usov (2009): Information Modelling and Simulation in Large Interdependent Critical Infrastructures in IRRIIS. In: Proceedings of the 3rd International Workshop on Critical Information Infrastructures Security, CRITIS 2008, Rome, Italy, October 13–15, 2008, LNCS 5508, Springer, Berlin.

20

T. Becker et al.

J. -C. Laprie, K. Kanoun, M. Kaaˆniche (2007): Modelling Interdependencies Between the Electricity and Information Infrastructures. In: Proceedings of the Computer Safety, Reliability, and Security. 26th International Conference on SAFECOMP 2007, Nuremberg, Germany, September 18–21, 2007, LNCS 4680, Springer, Berlin. T. Liebich (2009): IFC 2x Edition 3. Model Implementation Guide. Version 2.0. AEC3 Ltd. Online from http://www.iai-tech.org, last access 13. 4. 2010. B. Meehan (2007): Empowering electric and gas utilities with GIS. Series on Case Studies in GIS, ESRI Press, Redlands, CA. P. Pederson, D. Dudenhoefer, S. Hartley, M. Permann (2006): Critical Infrastructure Interdependency Modeling. A Survey of U.S. and International Research. Published by Idaho National Laboratory, US Department of Energy (INL/EXT-06-11464). R. Setola, S. Geretshuber (eds.) (2008): Proceedings of the 3rd International Workshop on Critical Information Infrastructures Security, CRITIS 2008, Rome, Italy, October 13–15, 2008, LNCS 5508, Springer, Berlin. W. J. Tolone, D. Wilson, A. Raja, W. Xiang, H. Hao, S. Phelps, E. W. Johnson (2004): Critical Infrastructure Integration Modeling and Simulation. In: Proceedings of the 2nd Symposium on Intelligence and Security Informatics, ISI 2004, Tucson, AZ, USA, June 10–11, 2004, LNCS 3073, Springer, Berlin. A. Usov, C. Beyel (2008): Simulating Interdependent Critical Infrastructures with SimCIP. In: ECN und European CIIP Newsletter. Online available at http://www.irriis.org/ecn/SimCIP_ Usov_Beyel.pdf, last access 27. 4. 2010.

Modeling Space by Stereographic Rejection W.L. (Pim) Bil

Abstract 3D geo-information analyses topological and metrical relationships between spatial objects. This analysis needs a suitable representation of the threedimensional world. This paper proposes to use the 4D unit sphere as a model. In essence this model is already present in mathematical theories like Lie sphere geometry, Moebius geometry and Geometric Algebra. The forementioned theories use the stereographic projection implicitely to build the model. This paper explicitely uses this geometric transformation to introduce the model as simply as possible following both an intuitive geometric and a formal algebraic self-contained way. The calculation in a CAD-environment of 3D Voronoi cells around given 3D points gives a straightforward example of the topological and metrical capabilities of this model. The addition of geometrical meaningful algebraic operations to the model will increase its computational power.

1 Introduction 1.1

Statement and Significance of the Problem

3D geo-information analyses topological and metrical relationships between spatial objects. A first order analysis is a linear one: planar forms represent the spatial objects and linear algebra describes the relations. This paper proposes a second order approach: the addition of spherical forms to the representation of spatial objects while retaining efficient computations. The stereographic projection is the geometric transformation that does the work. Earlier work described the linearized representation of 3D-spheres on the 4D unit sphere to facilitate geodetic calculations (Bil 1992).

W.L. (Pim) Bil Gemeente Amstelveen, Amstelveen, The Netherlands e-mail: [email protected]


21

22

1.2

W.L. (Pim) Bil

Sources

A representation of the stereographic projection on the unit sphere in terms of homogeneous coordinates was found by Hestenes and Sobczyk in a study of the spinor representation of the conformal group (Hestenes and Sobczyk 1984, Sect. 8.3). For efficient computation in manipulating geometric objects Hestenes et al. developed a unified algebraic framework and mathematical tools called Geometric Algebra (Hestenes et al. 1999). Dorst et al. implemented Geometric Algebra as a high-level language for geometric programming (Dorst et al. 2007). Lie sphere geometry represents points as spheres with radius zero (Blaschke 1929; Cecil 1992). The idea to use stereographic projection for the determination of Voronoi cells from the convex hull is found in Brown (1979). The construction of the Voronoi Diagram in the model is pointed out in Dorst et al. (2007). Ledoux mentioned the problem of using a dynamic Voronoi Diagram in a geographical information system (Ledoux 2008).

1.3

Historic Notes

The use of the stereographic projection to perform spherical calculations and to make maps has a long history. Already the Greek astronomer Hipparchus (180–128 BC) was aware of the projection. Around the first century the Alexandrian Claudius Ptolemaeus wrote on it in the book Planisphaerium. The construction of ancient Greek and medieval arab astrolabes uses its properties. Around the millennium in the Arab world Al-Bı¯ru¯nı¯ applied the stereographic projection to the making of maps. In 1587 the cartographer Gerhard Mercator invented conformality in his mappings of the eastern and western half spheres in stereographic projection. The Jesuit Aguilonius (1566–1617) is the first to use the name stereographic projection in his books on optics (Rosenfeld 1988; Grafarend and Krumm 2006).

1.4

Overview

For a first introduction the standard description of the (conformal) model in the literature on Geometric Algebra is too abstract, defining algebraic operations on coordinate free geometric objects on a null cone in five dimensions. The entrance to the model on the 4D unit sphere by the stereographic projection with just linear algebra seems simpler and because of the spherical symmetry has also an aesthetic appeal. The central thought is the idea to consider 3D linear space to be the stereographic projection of a four-dimensional unit sphere. This elementary introduction to the model is both on an analytic and a synthetic level. In the first part the analysis of drawings stimulates geometric intuition.

Modeling Space by Stereographic Rejection

23

With some concepts from projective and inversive geometry the stereographic projection of points, planes, and real and imaginary spheres is studied. One can look at the figures in two ways: as plane drawings on paper of the stereographic projection of a circle on a line, but also more in general as a mnemonic device showing the properties of the stereographic projection of a sphere on a linear space one dimension lower. In the second part vectors (i.e. coordinates) describe the position of geometric objects in the vector spaces Rn and Rnþ1, and linear algebra deduces the metrical and topological properties of the objects and their relations. In the end the construction of Voronoi cells around points in space exemplifies the use of the model.

2 Mathematical Preliminaries The explanation relies in the first place on the intuitive interpretation of drawings. Therefore this is not a mathematical text, although the goal is to be as precise as possible and to get acquainted with mathematical concepts. In mathematical terms the stereographic projection is an inversion, i.e. a special projective transformation. See for inversive geometrical concepts for instance (Pedoe 1979) and (Brannan et al. 1999). The core of projective geometry is the duality between point and plane, and the crux of inversive geometry is the concept of harmonic conjugacy. Figure 1 illustrates in the plane some inversive concepts to be used. For example the points P1 and I are inverses, i.e. OI OP1 ¼ 1. Indeed the triangles OTP1 and IOT are similar.

T

P3 r p3

P1

pola

1 I O

po

P2

p2

r p1

a pol

lar

Fig. 1 Construction of pole and polar with regard to the unit sphere

24

W.L. (Pim) Bil

OI OT 1 So OT ¼ OP , OI 1 ¼ OP1 , OI OP1 ¼ 1. In the following the inversive 1 concepts pole and polar play a central role. A point P is the pole of a linear space (its polar) with regard to a conic. Recipes for the geometric construction of the polar of a point Pi with regard to the unit sphere are: l

l

l

P1 outside the unit sphere: construct the tangents from P1 to the unit sphere. The polar is the linear space containing the tangent points on the unit sphere P2 inside the unit sphere: consider P2 as bundle of polars. Each polar has a pole. The polar of P2 is the collection of these poles P3 on the unit sphere: the polar is the tangent to P3

To keep the figure readable tangents are only drawn between the pole and the point of tangency. Homogeneous coordinates are suited to describe the duality between point and plane. See for this concept an elementary text on projective geometry, for instance (Ayres 1967). Homogeneous coordinates of a point or plane are the set of numbers that fulfill a homogeneous linear equation. Introduction of extra dimensions (coordinates) makes equations homogeneous. Consider for instance the equation vi au1 þ bu2 ¼ c. With the substitution ui ¼ ;i ¼ 1;2 this equation becomes homov 3 v1 v2 geneous: a þ b ¼ c , av1 þ bv2 cv3 ¼ 0, and after substitution the quadra 2 2 v3 v3 tic equation u1 2 þ u2 2 ¼ 1 becomes vv13 þ vv23 ¼ 1 , v1 2 þ v2 2 v3 2 ¼ 0. If a triple of coordinates ðv1 ; v1 ; v3 Þ fulfils a homogeneous equation, so does the triple ðlv1 ; lv2 ; lv3 Þ; l 2 R. The equivalence class of the relation ðlv1 ; lv2 ; lv3 Þ ðv1 ; v2 ; v3 Þ; l 6¼ 0, with not every vi ¼ 0, constitutes the homogeneous coordinates of a point or plane. The symbolism ~ x2 ¼~ x ~ x ¼ k~ xk2 stems from Geometric Algebra: the square denotes the inner product of a vector with itself and is equal to the square of the length of the vector.

3 Geometric Analysis of the Stereographic Rejection To get a better grasp on concepts such as stereographic projection, pole and polar it is instructive to first study some drawings. The notation Sn signifies the n-dimensional unit sphere in the (n þ 1)-dimensional vector space Rnþ1.

3.1

Points

Consider the south pole of the unit sphere Sn to be a bundle of lines (see Fig. 2). Every vector ~ x in the vector space Rn is on a line in the bundle. This line contains


25

Fig. 2 The stereographic x rejection U on Sn of vector ~ in Rn

U

o Sn

Rn

→

x

→

0

∞

another point U of Sn. Point U is called the stereographic rejection1 of ~ x, and ~ x the x on Sn other points of Rnþ1 are on stereographic projection of U.2 Beside the point ~ the line in the bundle through the center of projection and point ~ x in Rn. ~ x is called the stereographic projection of all these other points. The north pole is the stereographic rejection of the origin. The center of projection is a representation of infinity, called the point at infinity and denoted by 1: the greater the distance of a point in Rn to the origin, the closer the stereographic rejection of the point on the unit sphere to the center of projection. 1 0 0 1 ~ ~ 0 0 C B B C In homogeneous coordinates: o ¼ @ 1 A, 1 ¼ @ 1 A. 1 1 Note that the lines on the polar of the center of projection, parallel to Rn, have neither an intersection with Rn nor a second intersection point with Sn.

3.2

Real Spheres

A n-sphere in Rn is stereographically rejected on a (n þ1)-sphere on Sn that is part of a (n þ1)-plane in Rnþ1 (see Fig. 3). The stereographic rejections of points inside the n-sphere are all on the same side of the (n þ1)-plane. This (n þ1)-plane is polar of a point P. Pole P is geometrically determined as the intersection point of all tangent planes to Sn at the rejection of the n-sphere. The stereographic projection of P gives back the center of the n-sphere in Rn. All points outside Sn can be considered as 1

Here rejection is used in the sense of ‘back’ projection. In Geometric Algebra the rejection of a vector has a different meaning: the component complementary to its projection on another vector. 2 For a better symbolic discrimination of the elements in Rnþ1 in stead of the letter X the letter U is used for a general element, and the letter P denotes a pole. Consequently coordinates in Rnþ1 are denoted by ui .

26

W.L. (Pim) Bil

Fig. 3 Stereographic rejection to Sn of a sphere in Rn around m ~

P

U1

Rn

→ x1

→ m

∞

→ x2

U2

Sn

poles, corresponding to polars cutting Sn in (nþ1)-spheres that are stereographic rejections of n-spheres.

3.3

Planes

If point P is on the polar of 1, it can not be the stereographic rejection of a point in Rn, as shown in Sect. 3.1. The center of projection 1 is on the polar. The polar is a (nþ1)-plane and its intersection with Rn a n-plane. Thus pole P dually represents a n-plane. Figure 4 is a two dimensional cross section through the axis of the unit n is the normal vector of the plane. The sphere. Figure 5a is a section through Rn. ~ equation of the plane is ~ n ~ x¼~ n~ n. From the inversive relation, shown also in Fig. 1, follows: k~ nkkl~ nk ¼ 1 , l ¼ ~n12 . 0 1 ~ n So the pole has coordinates @ ~ n2 A . 1 The more the distance of pole P to the origin increases, the more ~ n tends to ~ 0 (Fig. 5b) and in the end the polar will contain the axis of the unit sphere. The normal vector of the plane in Rn will then also be the normal vector of the plane in Rnþ1.

3.4

Position of Pole and Length of Radius

A n-point is the stereographic projection of all poles corresponding to n-spheres around this n-point (see Fig. 6). The greater the radius of the n-sphere, the greater the distance of its corresponding pole to the stereographic rejection of the n-point on Sn. The n-sphere with radius zero corresponds to the stereographic rejection of the n-point on Sn, equal to the pole of the tangent. So all points in Rn, considered as circles with radius zero, are stereographic projected on S.


27

p

Fig. 4 Pole P on the polar of 1 dually represents a plane in Rn

Sn →

n

→

Rn

0

2

→ →

n /n

p

∞

Fig. 5 (a) Cross section looking from the north pole. (b) The limiting case: pole P infinitely far from the origin, ~ n becoming ~ 0, polar p containing north and south pole

p

P

a

b Rn

Sn

P p

p →

0

n→

P

→

0

p

p P

Fig. 6 The poles corresponding to spheres around a fixed point with different positive radii

Rn

∞

Sn Rn

∞

3.5

Imaginary Spheres

Thus far points in Rnþ1 on and outside Sn are seen to represent points and spheres in Rn (see Fig. 7). What about points of Rnþ1 inside Sn? Point P is dually defined as

28

W.L. (Pim) Bil pola

r of P

PC–λ

PCm

PCλ

PCμ

P

→

m

Rn

C–λ Cm CλCμ

C–λCm Cλ

Cμ

Sn ∞

Fig. 7 Point P defined as a bundle of (n þ 1)-planes corresponding to a collection of real n-spheres Ci

extra dimension Ua

Rn

→

m C–λ Cm Cλ Cμ

r r

C–λ Cm Cλ

Cμ

Ub

Fig. 8 Intersections, represented in the figure as Ua – Ub, that the n-spheres in the collection shown in Fig. 7, have in common in the extra dimension

a bundle of planes cutting Sn in (n þ 1)-spheres, with the PCi as the poles of these planes, together forming the polar of P. The (n þ 1)-spheres on Sn stereographically project on n-spheres. Embedding Rn in Rnþ1 all the extended n-spheres intersect in points in Rnþ1 at a ~ of P (see Figs. 8 and 9).3 certain distance of the stereographic projection m

3

See also Sect. 4.5.

s

s

sp here

e radius

∞ planes

sitiv

spheres with imaginary radius

iu

po

i

ints)

sph

w

po

sp

re

s

radius ze ith r (po

eres with

ius sphere

th

w

he

3.6

rad

e ti v

o

Fig. 9 Points in Rnþ1 dually represent points, planes and real and imaginary spheres in Rn depending on their position with regard to the unit sphere

29

si


ra s w e ith positiv

d

Rnþ1 Representing Spheres in Rn

Summarizing our analysis of the polar representation of spheres in Rn as points in Rnþ1: l l l l l

The center of the stereographic projection represents infinity: 1 Points on the polar of 1represent planes Other points outside the unit sphere represent real spheres Other points on the unit sphere represent points Points inside the unit sphere represent imaginary spheres

4 Algebraic Synthesis of the Stereographic Rejection Practical applications need quantities. In the following sections linear algebra revisits the results found earlier and quantifies the relations between the coordinates of the points in the vector spaces Rn and Rnþ1.

4.1

Pole and Polar

The polar relationship can be expressed in terms of linear algebra as follows. Point A is on the unit sphere, so ~ a~ a ¼ 1 (see Fig. 10). If a point is on the tangent to A, then ~ x~ a is perpendicular to ~ a, i.e. ð~ x~ aÞ ~ a ¼ 0. So ~ x~ a¼~ a~ a ¼ 1. In R2: Let P be the polar of the line AB. P is on the tangent through A, * so ~ p~ a ¼ 1, and P is on the tangent through B, so also p b~ ¼ 1. Let X be ~ another point of the line AB. If ~ x¼~ a þ lðb ~ aÞ; l 2 R then ~ p ~ x¼~ p~ aþ ~ lð~ pb~ p~ aÞ ¼ 1. So the equation of the polar of P is ~ p ~ x ¼ 1.

30

W.L. (Pim) Bil

Fig. 10 Pole P and its polar ~ p ~ x¼1

→ →

p.

p .x >1

x= 1

→→

P

→→

p . x 1 and ~ p ~ x < 1.

4.2

Points

The line of projection from the south pole has equation (see Fig. 11): !! ! ! ~ ~ ~ x x 0 ~ ð1 lÞ~ x u ; l 2 R: ¼ þl ¼ l unþ1 0 0 1 Now ~ u 2 þ unþ1 2 ¼ 1, so ð1 lÞ2~ x 2 þ l2 ¼ 1 , 2 2~ x2 þ 2 ¼ 1: x 2 2l~ x 1 ¼ 0 , l1 ¼ x2 þ ~ l2 1 þ ~ 2 1 þ~ x2 This gives the south pole, and the other solution is: 2 2~ x2 2 ~ x 1 ¼ : l2 ¼ 2 1 þ~ x2 1 þ~ x2 Using this value gives for the coordinates of the stereographic rejection U of ~ x: 0

2 1 0 1 ~ x 1 2~ x ~ x 1 B C B 1 þ~ 1 þ~ x2 x2 C ~ u C B C B ¼B ¼ C: B C unþ1 @ A @ 1 ~ x2 A ~ x2 1 1 þ~ x2 1 þ~ x2


31

Fig. 11 The stereographic rejection in coordinates

Sn : → u 2 + u2n+1 = 1

un+1

Rn

→

x

→

u 1 ∞

Fig. 12 The vector representation of the stereographic projection

→ u un+1

un+1 → x

Rn

→

u

1

Sn ∞

In homogeneous coordinates this is4: 1 1 0 2~ x ~ x B C B 1 þ~ x2 C 1 ~ x2 C C B B B C C B C ¼ Sð~ x2 C B xÞ; B 1 ~ 2 B C C B B C 2 A 2 @ 1 þ~ x @ 1 þ~ x A 1 2 0

(1)

in which the function S : Rn ! Rnþ2 defined by (1) is the stereographic rejection expressed in homogeneous coordinates. ~ u See Fig. 12, the stereographic projection of 2 Rnþ1 on Rn is: unþ1

See Sect. 2: substitute ui ¼

4

vi and scale to get a suitable representation. vnþ2

32

W.L. (Pim) Bil

2 2~ x ~ x ~ u 1 þ~ x2 1 þ~ x2 ~ ¼ x¼ ¼ : 2 2 unþ1 þ 1 1 ~ x þ1 1 þ~ x2 1 þ~ x2

4.3

(2)

Real Spheres

~Þ2 ¼ r2 . In Rn the equation of a sphere around a point M is ð~ xm ~ u Using (2), substitution of ~ x¼ gives: 1 þ unþ1 2 ~ ~ 2~ m~ u u u2 ~ ¼ r2 , ~2 ¼ r 2 , m þm 1 þ unþ1 ð1 þ unþ1 Þ2 1 þ unþ1 1 unþ1 2 2~ m~ u ~2 ¼ r 2 , þm ð1 þ unþ1 Þð1 þ unþ1 Þ 1 þ unþ1 ð1 þ unþ1 Þð1 unþ1 Þ 2~ m~ u ~2 ¼ r 2 , þm ð1 þ unþ1 Þð1 þ unþ1 Þ 1 þ unþ1 2 ~ 1 r 2 unþ1 ¼ ~ m2 1 þ r 2 , 2~ m~ uþ m

(3)

~2 r2 1 m 2~ m~ u ~2 ¼ m ~ m ~2R þ unþ1 ¼ 1; with m ~ m2 1 þ r 2 ~ m2 1 þ r 2

(4)

1. According to Sect. 4 we read off from this equation of the and ~ u 2 þ u2 nþ1 ¼ 0 1 2~ m B C m2 1 þ r 2 C B ~ plane in Rnþ1 that B 2 C is pole of it. @ m ~ 1 r2 A ~ m2 1 þ r 2 The stereographic projection of this pole on Rn gives (2): 2~ m 2 ~ u * ~ m 1 þ r 2 ¼ m: ¼ 2 1 þ unþ1 ~ 1 r2 þ ~ m m2 1 þ r 2 ~ m2 1 þ r 2 ~Þ2 < r 2 , in the same way: Starting from the equation ð~ xm 2 ~2 1 þ r 2 : ~ 1 r2 unþ1 < m 2~ m~ uþ m

(5)

Equation (5) is always valid, because unþ1 þ 1 0. However, the sign of the equation corresponding to (4) is depending on the values of the parameters in the ~2 1 þ r 2 . expression m


4.4

33

Planes

Consider a plane in Rn with normal vector ~ n:~ n ~ x¼~ n 2 . The equation of the plane nþ1 in R containing stereographic rejected points is: ~ ~ u n ¼~ n2 , ~ n ~ x~ n 2 xnþ1 ¼ ~ n2 , 2 ~ u unþ1 ¼ 1: ~ n 1 þ unþ1 0 1 ! ~ n ~ 0 2 is on the (n þ 1)-plane. The pole is @ ~ n A, and the center of projection 1 1 ~ n

4.5

Position of Pole and Length of Radius

It has been made clear the position of the pole depends on the radius of the n-sphere (see Fig. 6). The homogeneous coordinates of the pole of the (nþ1)-plane that contains the stereographic rejection are deduced from (4): 1 0 2~ m 1 0 2 2 C B ~ 2~ m B m 1 þr C C C B 2 B ~2 1 r 2 C @ m ~ 1 r2 A B m C B @ ~ m2 1 þ r 2 A ~ m2 1 þ r 2 1 0

~ m

1

1 0 B ~ 2C 0 C 1 B1 m ~ 1 C B C B B 2 C r 2 @ 1 A ¼ Sðm ~Þ r2 1: C 2 B 2 @1 þ m 1 ~2 A 2

(6)

In (6) the equivalence class of homogeneous coordinates of the pole is represented by the coordinate tuple that encodes the stereographic rejection of the midpoint [see (1)]. This way the homogeneous vector of the pole, a (n þ 2)-vector, carries the quantitative information about the corresponding n-sphere. Because on the model no metric and geometric algebra operations are defined, it is not possible to exploit this result here.

4.6

Imaginary Spheres

Having found an algebraic relation between points on Sn and points in Rn, and between points outside Sn and n-planes and n-spheres with positive radius, motivated by the geometric analysis in Sect. 3.5 the next search is for the algebraic

34

W.L. (Pim) Bil

Fig. 13 The simplest case: an imaginary 1-sphere (point pair) UaUb at imaginary distance r from M is part of the imaginary 2-sphere Cl with midpoint L and radius R containing the 1-sphere (point pair) x1 x2 on the line n

extra dimension

R x1

n

L

r M

m-l Cl

Ua

x2

n

r Ub

relationship between points inside Sn and n-spheres with an imaginary radius (see Fig. 9). Consider the line n in R1 as simplest case (see Figs. 8 and 13). Embed this line in a space one dimension higher. If a midpoint M is given, determine the points with orthogonal distance r to M: the two-sphere with imaginary radius r. For all other point on 2D line n: draw the two-sphere (circle) containing the imaginary points Ua and Ub. This means on line n there are points x1 and x2 fulfilling the equation: ðm xÞ2 ¼ R2 ¼ ðm lÞ2 þ r 2 .5 Take such a one-sphere with point L, having coordinate value l, as center. According to (3) the equation of the two-line that is the stereographic rejection of the one-sphere is: 2lu1 þ ððl2 ðl mÞ2 þ r2 Þ 1Þu2 ¼ ðl mÞ2 þ r 2 l2 1:

(7)

The intersection point of this two-line with the two-line that is corresponding with the two-sphere through the imaginary points Ua and Ub with center l gives 2m 1 m2 r2 and u ¼ . This expression is indeas coordinates u1 ¼ 2 1 þ m2 þ r2 1 þ m2 þ r2 pendent of l. Repeat the construction in Rn by considering the points related to m ~ by * lÞ2 þ r 2 fulfilling the equation equivalent to (7): the equation: ð~ m ~ xÞ2 ¼ ðm ~ ~Þ2 þ r 2 Þ 1Þunþ1 ¼ ðð~ ~Þ2 þ r2 Þ2 ~ 2~ l~ u þ ð~ l 2 ð~ lm lm l 2 1:

(8)

0

1 2~ m B1 þ m ~2 þ r2 C C is true, for P fulfils (7), The conjecture all planes contain P ¼ B @1 m ~2 r2 A ~2 þ r2 1þm * n independent of l . So if a sphere Ci in R cuts the imaginary sphere with center M, P is on the plane in Rnþ1 that contains the stereographic rejection of Ci (see also 5

Historic aside: this is in fact the ancient construction of the mean proportional from Proposition 13 of book VI in The Elements of Euclid.


35

Figs. 7 and 8). The representant of the homogeneous coordinates of pole P [compare this with (6)] is: 1 0 2~ m 1 0 B 1þm 2~ m ~2 þ r 2 C C B C B B C ~2 r 2 C @ 1 m ~2 r 2 A B 1m C B 2 @ 1þm ~ þ r2 A ~2 þ r 2 1þm 1 0

~ m

1

1 0 B ~ 2C 0 C 1 B1 m ~ 1 C C B B ~Þ ðir Þ2 1 B 2 C þ r2 @ 1 A ¼ Sðm C 2 B 2 @1 þ m 1 ~2 A 2 (nþ1)-point P lies inside Sn. Indeed, given the fact that ~2 þ r2 Þ2 ð1 m ~2 r 2 Þ2 ¼ ð2~ ð1 þ m m2 þ 2r 2 Þð1 þ 1Þ ¼ 4~ m2 þ 4r 2 ; ~2 r2 Þ2 þ 4r2 ¼ ð1 þ m ~2 þ r2 Þ2 . it follows that 4~ m2 þ ð1 m 2 2 ~2 r 2 2~ m 1m So, because 4r2 > 0 , P2 ¼ þ < 1. ~2 þ r 2 ~2 þ r2 1þm 1þm ~, for from (2) follows: Again this pole P is projected on m 2~ m ~2 þ r 2 ¼ m 1þm ~: ~2 r2 1m 1 ~2 þ r2 1þm

5 Example: Spatial Partitioning by Voronoi Cells 5.1

Introduction

The calculation of a partition of nD with the help of the topological and metrical properties of its rejection on the (nþ1)D unit sphere shows the model at work. In a number of random points ~ p 2 S Rn a quantity is measured. A natural p. approach is to attribute to a point ~ q 2 Rn the measurement of the nearest point ~ The Voronoi cell is the region of points that are closest to ~ p. The mathematical definition of the Voronoi cell is: Vp ¼ f~ x 2 Rn j8~ q 2 S : jj~ x~ p jj jj~ x~ qjjg:

36

5.2

W.L. (Pim) Bil

Analysis of the Voronoi Diagram in Two Dimensions

A planar analysis shows this to be a spherical construction. The black points are randomly given (see Fig. 14). The edges of the Voronoi cells around the black points consist of points with equal distance to two nearest black points. The branch points (open circles in the drawing) consist of points with equal distance to three nearest black points. That is to say the branch point is the center of the circle to the three nearest neighbouring black points. These three black points form triangle with the property that no other black point lies inside the circumcircle, a Delaunay triangle. The plane is triangulated. The edges connecting the midpoints of the triangles around a black point make up the convex Voronoi cell (Fig. 15).

Fig. 14 Voronoi Diagram (of white points) and Delaunay triangulation (of black points)

Fig. 15 Detail around a black given point. Five Delaunay triangles meet in this point. The circumcenters of the triangles are the vertices of the convex Voronoi cell around the point


5.3

37

The 2D Voronoi Diagram on the 3D Unit Sphere

See the scheme in Fig. 16. The stereographic rejections of the given points are on the 3D unit sphere. A Delaunay triangle of these points has the topological property that its circumcircle does not contain another given point. This property is equivalent to the topological property on the unit sphere that all stereographic rejections of the given points lie on the same side of the plane formed by the three points. In other words the Delaunay triangulation in the plane corresponds to the convex hull on the 3D unit sphere. The pole of the plane formed by the three points on the 3D sphere gives the stereographic rejection of the center of the circumcircle of these points. The mouthful formulation of the conclusion of the plane analysis is: the Voronoi cell of a given point is the convex hull of the stereographic projection of the poles of the facets around the stereographic rejection of the given point of the convex hull of the stereographic rejection of all given points.

5.4

Calculation of the 3D Voronoi Diagram

The abovementioned formulation is valid for the creation of a Voronoi Diagram in all dimensions. The following focuses on the construction in R3. The algorithm for the construction of Voronoi cells in R3 is: 1. Input random points in space 2. Stereographically reject these points on the 4D unit sphere 3. Calculate the convex hull of these points in 4D. In non degenerate cases the convex hull consists of n facets of four vertices. The number n depends on the configuration of the given points in space 4

Sn 2

3 2 1

Rn

5

1

1

5 4

4

3 1

5

1

5

3 2 2 4

3 2

3

3 4

∞

4

Fig. 16 The relation between the Voronoi and Delaunay tessellation, both in Rn and in Rnþ1

38

W.L. (Pim) Bil

4. Calculate the poles of the n planes incident with these facets 5. The stereographic projections of these n poles are the midpoints of the circumspheres of four given points (tetraeders), building blocks of the Voronoi cell around a given point in 3D 6. Calculate the convex hull of the stereographic projection of these n points around the given point in 3D. This gives the 3D facets of the Voronoi cell around a given 3D point

5.5

Neighbouring 3D Voronoi Cells

Neigbouring 3D Voronoi cells share a 2D facet. To find the connection of a Voronoi cell, first compare for all facets the number of vertices between all other calculated facets. In case of equality next compare the equality of the vertices of the facet with equal number of vertices. The order of the vertices of the matching facet is reversed. The connectivity of the Voronoi cells follows.

5.6

Moving Points

See Fig. 17 and also Sect. 4.1. If the given points move, so do their stereographic rejections, the facets/polars they are part of, the corresponding poles and the 3D Voronoi vertices. Sometimes the topology of the configuration changes. Having P1

P3

P2

U2

U3

U4

U1

Fig. 17 Check of consistency on S1: the vertices U1, U2, U3 and U4, on the polars of P1 and P3, disturb the convex hull structure

P4 U5 P5


39

available the positions of the moving points in real time, the following test checks the consistency of the configuration. Given the fact the configuration must be a convex hull, no point on the unit sphere should lie outside the polars formed by the facets of this hull. So for all poles all coordinate vectors of points Ui on the unit sphere ought to satisfy either the equations ~ pj ~ ui 1, or ~ pj ~ ui 1. For example in Fig. 17. the movements of points U2 and U3 have distorted the original convex hull U1-U2-U3-U4-U5-U1. For pole P1 the sign of points U3 is different from U4 and U5, and for pole P3 the sign of U2 is different from the points U1 and U5. In this case not only the position of the points and poles has to be updated, but also (part of) the topology.

6 Implementation As proof of principle this theory has been implemented in software.

6.1

CAD

Given input is an ASCII-file with coordinates of points. The given points are then represented in a 3D Microstation designfile. Visual Basic for Applications for Microstation is the programming environment. The coordinates are read from the designfile and stored in the array of vertex structures that is kept in memory to speed up calculations. All coordinates are translated and scaled towards the 3D unit sphere. In this process the maximum distance of the vertices to the origin is determined. After the coordinates of the given points are read the 3D coordinates of an icosahedron are added to enclose all the given points. The 12 vertices ð0; tRmax ; Rmax Þ; ð tRmax ;0; Rmax Þ and ð Rmax ; of the icosahedron are: pffiffi 5þ1 tRmax ;0Þ; with t ¼ 2 the divine proportion (Coxeter 1969). The addition of the south pole of the 4D unit sphere to the stereographically rejected 3D points closes the set of 4D points on the sphere. Pointers from within the vertex structures address the Voronoi coordinates (derived from the data in the array of facet structures).

6.2

Convex Hull Software

An introduction into the convex hull computation is given in (de Berg et al. 1998). In the application the calculations of the convex hull are done using QHull. Information on QHull can be found on http://www.qhull.org/. The input for the qconvex executable is formed by writing 4D or 3D coordinates from the Microstation 3D designfile to an ASCII-file. After execution of the command the outputted ASCII-file is read. The following command lines were used:

40

W.L. (Pim) Bil

Fig. 18 CAD drawing of a Voronoi cell around a 3D point

1. qconvex o to calculate the 3D and 4D convex hulls. It outputs on the first line the dimension, on the second line the number of vertices, facets and ridges, then lines with vertex coordinates and finally the facet lines. Each facet line starts with the number of vertices. The vertex indices follow. In general for the convex hull on the 4D unit sphere the facets are simplices and made up by four vertices, corresponding with the 3D counterpart: four vertices determine a sphere in 3D. If more vertices are encountered this signals a degenerate configuration of the points and calculation measures can be taken, for instance by a little shift of a point concerned. 2. qconvex FN to list the number of facets on the first line and the neighbouring facet indices for each facet on the next lines. The line starts with the number of neighbouring facets (Fig. 18).

7 Conclusion, Caveat and Developments The paper proposes to model space on the four-dimensional unit sphere, described by five homogeneous coordinates. Space is considered to be the stereographic projection of the 4D unit sphere. Points on the 4D unit sphere thus represent 3D points, i.e. spheres with radius zero. The 4D center of projection is the representation of the point at infinity and on its polar lie the 4D points that represent 3D planes. Points outside the 4D unit sphere dually represent 3D spheres with positive radius, and 4D points inside represent 3D spheres with imaginary radius, which are given a clear geometric interpretation. The expression of the 4D points in 5D homogeneous coordinates contains quantitative information on the center and radius of the dually represented 3D spheres and planes. The first part of the paper introduces the concepts in an informal way analyzing the geometry in an intuitive manner. Next the conjectures are deduced in explicit formulae of elementary linear algebra. As a straightforward example, useful in geographical information systems, Voronoi cells around given points in space are determined from the convex hull in the model.


41

As a proof of principle the theory has been implemented in software. Because the main objective of the author is to introduce the model, and because he is not fully aware of the state-of-the-art of the construction of Voronoi Diagrams and use of data structures, the software was not developed to the point it processes dynamic input. Nevertheless algorithms to calculate the connectivity of the Voronoi Diagram and to validate the configuration were indicated. Microstation offers the possibility for event-driven programming. So it is possible to monitor the position of the given points and calculate the stereographic rejections of the given points and values depending on them, in real time. The validation test of the configuration of the Voronoi cells could trigger the recalculation of the topology while running a program. Although, as we have seen in the example of the calculation of the Voronoi Diagram, spheres and planes are represented on an equal footing in the model and we have linear algebra at our disposal, the model is not yet fully operational. At present the obtained methods are of limited use, for geometric meaningful algebraic operations are still lacking. To have the model at one’s disposal without operations defined on it, is like having hardware without software. Now that 3D space is beamed up along the rays of the stereographic projection to the four-dimensional unit sphere “one must so to speak throw away the ladder, after he has climbed up on it”6 and wander around in this copy of the world. Geometric Algebra provides a unified coordinate-free algebraic framework for both multidimensional geometric objects and geometric operations in this model. The conformal model of Geometric Algebra goes one dimension up and models the 4D unit sphere in 5D as a null cone, isometrically embedding 3D. As an algebra Geometric Algebra is closed, i.e. every geometric product of (geometric) elements gives another element in the algebra. A main advantage is that only one formula is sufficient to describe a geometric situation. No “special case” processing is needed. For instance two spheres will always intersect. Imaginary spheres have, as is shown in this paper, a clear geometric interpretation. To learn about the foundation of Geometric Algebra see (Hestenes and Sobczyk 1984) and about the conformal model see for instance (Hestenes et al. 1999), (Dorst et al. 2007), (Perwass 2008). Some communities have adapted the conformal model of Geometric Algebra as a computational tool. New applications of the model are frequently a topic at the International Workshops on Computer Graphics, Vision and Mathematics (GraVisMa) and the conferences on Applied Geometric Algebras in Computer Science and Engineering (AGACSE). It remains to be seen whether or not the GA conformal model of Euclidean 3D geometry has killer applications in 3D geo-information.

6

Satz 6.54 from the Tractatus logico-philosophicus of Ludwig Wittgenstein.

42

W.L. (Pim) Bil

References Ayres, F. (1967) Theory and Problems of Projective Geometry, Schaum’s Outline Series, McGraw-Hill, New York Bil, W.L. (1992) Sectie en Projectie, Nederlands Geodetisch Tijdschrift Geodesia, 10:405–411 Blaschke, W. (1929) Vorlesungen € uber Differentialgeometrie III, Springer, Berlin Brannan, D.A., Matthew, F.E., Gray, J. (1999) Geometry, Cambridge University Press, Cambridge Brown, K.Q. (1979) Voronoi diagrams from convex hulls, Information Processing Letters, 9:223–228 Cecil, T. (1992) Lie Sphere Geometry, Springer, New York Coxeter, H.S.M. (1969) Introduction to Geometry: De Divina Proportione, John Wiley & Sons, New York de Berg, M., van Kreveld, M., Overmars, M. and Schwarzkopf, O. (1998) Computational Geometry: Algorithms and Applications, 2nd edn, Springer, Berlin, Germany Dorst, L., Fontijne, D. Mann, S. (2007) Geometric Algebra for Computer Science, An Object Oriented Approach to Geometry, Morgan Kaufmann, Massachutas, USA Grafarend, E.W., Krumm, F.W. (2006) Map Projections, p. 72: Historical Aside: Stereographic Projection, Springer, New York Hestenes, D., Sobczyk, G. (1984) Clifford Algebra to Geometric Calculus, Reidel, Dordrecht Hestenes, D., Li, H., Rockwood, A. (1999) A unified algebraic framework for classical geometry: (1) A Unified Algebraic Approach for Classical Geometries. (2) Generalized Homogeneous Coordinates for Computational Geometry. (3) Spherical Conformal Geometry with Geometric Algebra. (4) A Universal Model for Conformal Geometries of Euclidean, Spherical and Double-Hyperbolic Spaces, in: Sommer, G. (ed), Geometric Computing with Clifford Algebra, Springer, London Ledoux, H. (2008) The Kinetic 3D Voronoi Diagram: A Tool for Simulating Environmental Processes, in: Oosterom, P.V., Zlatanova, S., Penninga, F., and Fendel E. (eds): Advances in 3D Geo Information Systems, Proceedings of the 2nd International Workshop on 3D Geoinformation, December 12–14, 2007, Delft, The Netherlands, Lecture Notes in Geoinformation and Cartography, Springer, pp. 361–380 Pedoe, D. (1979) Circles, a Mathematical View, Dover, New York Perwass, C.B.U. (2008) Geometric Algebra with Applications in Engineering, Springer, Berlin Rosenfeld, B.A. (1988) A History of Non-Euclidean Geometry, pp. 121–130: Stereographic Projection, Springer, New York

Rapid Modelling of Complex Building Interiors Pawel Boguslawski and Christopher Gold

Abstract Great progress has been made on building exterior modelling in recent years, largely driven by the availability of laser scanning techniques. However, the complementary modelling of building interiors has been handicapped both by the limited availability of data and by the limited analytic ability of available 3D data structures. Earlier papers of ours have discussed our progress in developing an appropriate data structure: this paper reports our final results, and demonstrates their feasibility with the modelling of two complex, linked buildings at the University of Glamorgan.

1 Introduction This paper completes the discussion of the development of a new data structure for full 3D building modelling, complete with interior volume elements and dual-graph connectivity between rooms. The initial paper (Boguslawski et al. 2009) gave the background of earlier structures, including the Quad-Edge (QE) of Guibas and Stolfi (1985) and the Augmented Quad-Edge (AQE) of Ledoux and Gold (2007). The first provided a 2D primal/dual structure on the surface of a simple shell, while the second gave a 3D primal/dual structure for cell complexes, with the dual graph providing the connectivity between the volume elements. Unfortunately the AQE, while providing the desired connectivity, did not have easy incremental construction operations, so (Boguslawski and Gold 2009) proposed an atomic element of two permanently linked

P. Boguslawski (*) Department of Computing and Mathematics, University of Glamorgan, Wales, UK e-mail: [email protected] C. Gold Department of Computing and Mathematics, University of Glamorgan, Wales, UK and Department of Geoinformatics, Universiti Teknologi Malaysia (UTM), Johor Bahru, Johor, Malaysia e-mail: [email protected]


43

44

P. Boguslawski and C. Gold

half-edges (the Dual Half-Edge or DHE), one in the primal space and a matching one in the dual space. It also proposed a construction mechanism (‘Cardboard and Tape’) where individual walls were constructed with DHEs and then ‘taped’ together. The second paper (Boguslawski et al. 2010a) elaborated on this, showing the four ways of combining a pair of DHEs and discussing the requirements for a set of ‘Euler Operators’ (EOs) that would work for cell complexes and not just individual shells. As EOs operate on individual edges, a system that operated on cell complexes (and not just the single shell of traditional EOs) needed to be able to connect both primal and dual edges simultaneously in order to preserve the dual navigation framework. The paper demonstrated that this was possible if the global model (such as a building) always had an exterior shell separating the building from ‘the rest of the world’. It described the construction of simple boxes, where faces and volumes were constructed automatically as edges were added, and also extensions where separate boxes were joined (sharing a common face but with individual volume elements and connectivity) and when they were merged (the face was removed and the two volume elements became one). There was a brief mention of escape route planning. Ledoux and Gold (2007), Boguslawski and Gold (2009, 2010a) all describe the work of other authors of manifold and non-manifold data structures, and they will not be repeated here. Briefly, only the AQE and the DHE permit full navigation within cell complexes and between volume elements. This paper completes the analysis by discussing manifold and non-manifold modelling in CAD, gives a clearer description of the DHE pointer structure, and discusses the assignment of attributes to the primal or dual manifestations of an edge or vertex. The DHE is compared with a related structure of Yamaguchi and Kimura (1995) and it is shown that not all non-manifold cases can be handled by their Coupling-Entity: in particular, cells connected by only a common vertex cannot be traversed successfully, as can be done with the ‘full’ model of the DHE. (Awkward non-manifold cases are important in advanced CAD systems, as they often arise as intermediate stages in the construction process.) The Coupling-Entity model is shown to have equivalent properties to the ‘simplified’ case of the DHE (where the dual graph is suppressed and the primal pointer that would have addressed it connects to an edge of the adjacent cell instead). Thus the full DHE is a more complete structure, justifying the inclusion (and extra storage) of the dual graph. In addition, it is shown to handle holes in cells as well as cavities. The paper concludes with a real-life example, two complex buildings connected by an overpass and containing some 1,300 cells when doorway and similar elements are included. Implementation of Dijkstra’s algorithm on the dual graph provides a real example of escape route planning, for which this project was originally developed.

2 Background: Manifold and Non-manifold Models Building exteriors are usually modelled as a set of connected tiles – often triangular. This ‘B-Rep’ (Boundary Representation) is similar to the TIN (Triangulated Irregular Network) models used in GIS for terrain modelling. As originally developed

Rapid Modelling of Complex Building Interiors

45

for CAD systems, the B-Rep and its set of associated Euler operators permits the incremental construction of complex single-shell entities: e.g. engine parts – or building exteriors (e.g. Kolbe et al. 2009). When this approach is applied to building interiors, problems arise. Building interiors may be thought of as a set of spaces (rooms or hallways) that are adjacent and connected: thus the model is no longer a single shell, or even a set of unconnected shells for the rooms, but a cellular complex whose boundaries are not a ‘twomanifold’, but something that requires a ‘non-manifold’ model, where, for example, more than two faces may meet at an edge. CAD developers have designed several systems to handle these cases, but they have been found to be complex to implement and they do not always handle every special case. This is particularly difficult during the construction process: Euler operators are designed to add one edge at a time, and this leaves temporary awkward structures that will usually disappear as the model is completed. Nevertheless, the connectivity of these cases must be fully defined in order to provide a stable and topologically complete data structure. Some non-manifold models are presented in Fig. 1. Two shells joined by a shared edge (Fig. 1a) or vertex (Fig. 1b) are typical examples. In two-manifold models an edge is coincident with exactly two faces, but in non-manifold models more than two faces can be joined to an edge. In Fig. 1b example there is one common vertex for two shells; edges sharing this vertex create two separate cycles (one for each cell) which is not allowed in two-manifolds. There are more examples: open volumes, mixture of shells, faces, separate edges and vertices, etc. Several of these non-manifold CAD models have been described in our previous papers (Boguslawski and Gold 2009, 2010a). In general they are complex and difficult to implement, and may not have all the properties one might desire. In our project we are particularly interested in escape route planning, which requires both volume entities (rooms) and the dual graph (connectivity between rooms). While these can always be obtained from the primal (geometric) graph describing the cell complex, this is an awkward way to proceed when they are of key interest, with in some cases rapidly changing attributes. No previous models included the dual graph as part of the structure [as is achieved in 2D, for example, by the Quad-Edge (QE) structure of Guibas and Stolfi (1985)]. In addition, we found it useful to have a well defined and permanent ‘outside’ or exterior shell, corresponding to the exterior building models obtained from laser scanning methods.

Fig. 1 Examples of non-manifold models

46


3 The Dual Half-Edge Data Structure As described in our earlier papers we found that the most satisfactory approach was to incorporate both the primal (geometric) and dual graph structures within our basic model, following the inspiration of Guibas and Stolfi (1985), the half-edge of M€antyl€a (1988) and the AQE of Ledoux and Gold (2007). It uses the half-edges to represent each polyhedron, but like the AQE adds the dual to each element. Thus we have pairs of half-edges, one in primal space and one in dual space, which are permanently linked together. Each half-edge has pointers: to a vertex; to the paired half-edge that forms the opposite side of the edge; to the next half edge around the associated vertex; and to the next half edge around its own face – so the primal part contains a loop pointer around the face of a single cell and the dual part contains a pointer around the face in dual space – which is equivalent to a loop around an edge in the primal space (see Fig. 2). This simplifies the overall structure considerably. It is interesting to note that the original half-edge structure consists of twinned pairs of half-edges in 2D, and the same would be true for a model of the dual. The Quad-Edge combines these two into an element containing four half-edges, with improved navigation features. The DHE splits these four half-edges in a different way: one primal and one dual in each element. This preserves the two-manifold navigation properties of the QE and the easy navigation between cells of the AQE – while allowing flexible ‘twinning’ of half-edges in the construction process. Thus each half edge may be linked to a single other half edge, forming part of a simple shell, as with b-rep models. To access adjacent shells one goes to the dual of one of these edges and moves around the dual face until the adjacent primal shell is found.

Fig. 2 DHE pointer based data structure; primal graph (solid lines) is connected permanently with the dual graph (dashed lines); he – original half-edge; S, NV, NF, D, V – pointers; Sym, NextV, NextF, Dual – navigation operators


47

There are no shells without an adjacent shell: the boundaries of the model are enclosed by an exterior shell, as described below. A complete primal (‘geometric’ or ‘building’) model involves vertices, edges, faces and volumes. Primal vertices and edges are equivalent to volumes and faces in the dual model. Primal faces and volumes are equivalent to edges and vertices in the dual model. Thus all elements may be represented by a graph structure consisting of vertices and edges – half in primal space and half in dual space. Observe that each element has a dual role: attributes may be assigned to its primal and/or dual meaning. In terms of storage, whether in a graph structure or a data base, there are only two atomic element types, vertices and edges. Both may have attribute information associated with their primal or dual roles, but only edges have topological pointer information: to the second half of the edge, to the dual edge, to the next edge around a vertex, to the next edge around a face, and to its associated vertex. Exterior shell – While this model is complete in the interior of a model, there could be problems at the exterior boundary. Each room has a simple QE type shell. The dual edge for each face connects the interior ‘room’ node with that of an adjacent room. If there is no adjacent room the dual graph structure – itself a collection of simple shells – will be incomplete, and some navigation operations could fail by ‘falling off the edge of the world’. To prevent this, and to make all the required Euler operators viable, a simple rule of the DHE is that there is always a single exterior shell – the ‘exterior’ or the ‘rest of the world’. Thus an exterior face of a room will have a dual edge connecting the ‘room’ vertex with a single universal ‘exterior’ vertex: all exterior faces will have dual edges connecting to this vertex. This property is particularly useful when joining initially-separate buildings (cell complexes). Without the external cell it is not possible to create dangling edges, and thus the incremental construction of models, as explained in the Sect. 5, would not be possible. Connection of two half-edges in primal space (geometry) does not cause a problem, but in the dual not-paired halves prohibit navigation within the model. However since the external cell is kept along with the internal cells, every half-edge has its counterpart in the adjacent cell: in the external or internal cell. We also note that a ‘simplified’ model may be developed from the DHE, where the dual graph is replaced by a pointer directly to an edge of each adjacent shell. This may suffice for some applications but, as described previously, volume and dual-edge entities are missing. In addition it fails to manage the case where two shells are connected only by a vertex – see Fig. 1. All other cases can be handled. If the application does not require these features then the simplified structure significantly reduces the storage space required. The published system closest to our way of thinking is that of Yamaguchi and Kimura (1995) who described a ‘coupling entity’ data structure to link the individual edges of each shell (room) so that navigation was possible both within and between rooms. However, we have found no example of large-scale models produced by their approach, although they claim that Euler operators may be developed on their structure. We have shown (Boguslawski and Gold 2010b) that their model is equivalent to our ‘simplified’ one – with the same limitations.

48


4 The Coupling Entity Data Structure and the Dual Half-Edge Yamaguchi et al. (1991) and Yamaguchi and Kimura (1995) introduced the coupling-entity data structure for representing non-manifold topology of threedimensional models. They are focused on the boundary representation. They introduced the feather – a new coupling-entity. There are two groups of pointers present in this structure: mate pointers and cyclic pointers. They describe relations between neighbouring entities. Three types of the mate pointers, shown in the Fig. 3a), points at another feathers in a model: fan mate (FM), blade mate (BM) and wedge mate (WM). Given the feather entity e, the fan mate of e is the feather on the matching face which shares just a vertex with e, the blade mate of e is the feather on the matching face which shares both a vertex and an edge with e, and the wedge mate of e is the feather on the same shell which shares an edge with e. The three cycles of feathers are defined: a disk – the next feather around a shared vertex, a loop – the next feather around a face, and a radial – the next feather around a shared edge. There is also a cycle orientation taken into account: clockwise (C) and counter-clockwise (CC). Thus six cyclic pointers are necessary – counter-clockwise: disk (CCD), loop (CCL), radial (CCR), and clockwise: disk (CD), loop (CL), radial (CR). These cycle pointers can be deduced from the mate pointers (1–6) as shown in Fig. 3b) (only counter-clockwise cycles are shown). Thus the full set of nine pointers can be reduced to three and mate pointers are used exclusively. CCL(e) = FM(BM(e))

(1)

CCR(e) = WM(BM(e))

(2)

CCD(e) = WM(BM(FM(e)))

(3)

Fig. 3 The feather entity: (a) mates; (b) cycles


49

CL(e) = BM(FM(e))

(4)

CR(e) = BM(WM(e))

(5)

CDðeÞ ¼ FMðBMðWMðeÞÞÞ

(6)

Because of this reduction, the functionality of the model is limited – no more than one disc cycle can be represented with the feather data structure (Yamaguchi and Kimura 1995). Thus some topological relations are not possible in a model, for example joining two objects at a vertex (see Fig. 4). There are two disc cycles that need to be joined (e1!e2!e3 and e4!e5!e6) – one disc cycle from each cell. But this is not possible with the mate pointers only – all mate pointers are already used to connect the neighbouring entities of a single cell. The solution of that problem that allows for this non-manifold case can be adding the disc cycle pointers back to the feather (CCD and CD). Thus navigation around shared vertex would be explicit and the loop and radial cycles could be still deduced using the mate pointers. The DHE is a more flexible data structure than the coupling-entity: non-manifold cases like the one presented in Fig. 4 can be modelled. It is also possible to simulate the feather. All the pointers can be derived from DHE. The wedge mate pointer WM (e) is explicitly represented with Sym (S pointer) (7); fan FM(e) and blade BM(e) mates can be represented as a sequence of basic DHE pointers or navigation operators [(8) and (9)]: FM(e) = e:Dual.Sym:Dual : e:D:S:D

(7)

BM(e) = e:Adjacent : e:D:NF :D:S

(8)

WM(e) = e:Sym : e:S

(9)

Fig. 4 Joining two cells at a shared vertex

50


The cycle pointers can be derived (10–15) from mate pointers, but there are also DHE pointers that can be used explicitly: CCL(e) = e:NextF : e:NF

(10)

CCR(e) = e:NextE : e:D:NF :D (e:Adjacent.Sym)

(11)

CCD(e) = e:NextV : e:NV

(12)

CL(e) = e:PrevF : e.D.S.NF .D.S

(13)

CR(e) = e:PrevE : e.S.D:NF .D.S (e.Sym.Adjacent)

(14)

CD(e) = e:PrevV : e.D:NV .D

(15)

It has been shown that models using the feather as a basic element can be simulated with the DHE data structure. It is also possible to show that a simplified version of DHE exists and this is an equivalent of the feather. Thus the feather is a subset of DHE. In the simplified version there is no dual and the NF pointer is removed from the structure – there are NV, S and D pointers left (the V pointer is not taken into consideration, because it does not have any topological function). Because the dual is no longer available in a model, the D pointer has a different meaning and does not point at the dual half-edge – it points at a half-edge in the adjacent face. This connection is the same as the fan mate connection in the feather. Equations (16)–(18) show the relations between two models. The mate pointers can be defined with DHE pointers: FM(e) ¼ e.Dual : e.D

(16)

BM(e) = e.Adjacent : e.D:NV .S

(17)

WM(e) = e:Sym : e.S

(18)

The cycle pointers can be derived from the mate pointers as shown in (1)–(6) replacing the mate pointers with DHE equivalents (16–18). Because the simplified version is an equivalent of the feather – a reverse translation is possible [(19)–(21)]: e.D = FM(e)

(19)

e:NV ¼ WM(BM(FM(e)))

(20)

e.S = WM(e)

(21)


51

Unfortunately the simplified DHE has similar limitations as the feather. Joining two cells at a shared vertex produce a model which is not valid. However all edges sharing the same vertex can be joined in one cycle using the NV pointer: e1!e2!e3!e4!e5!e6 (see Fig. 4). But navigation is not valid. For example the CL (clockwise around a face) cycle is defined in (4). The fan and blade mates can be replaced with the DHE equivalents [(19) and (20)]. Thus CL is defined as NV.S. The incapability to navigate around a face appears in two places: e3 and e6. The e3 edge will be used in the example. e3.NV.S points at the opposite end of e4. But in the valid face loop the next element is the other end of e1. This error is caused by defining the loop cycle using the disc cycle. Unfortunately it does not work for the case presented in Fig. 4 – the disc and loop cycles should be independent. However, in a cell complex decomposing 3D space where all the cells are joined by shared faces (eight cube cells connected in a big 2 2 2 cube complex), other methods can be used to navigate between edges sharing the same vertex. A single cell in a complex is two-manifold, and only one disk cycle is necessary to navigate around a shared vertex. Disk cycles in other cells are not joined together, but navigation through shared faces is possible, and access to these cycles is possible. It was shown in this section that the DHE data structure can be simplified to an equivalent of the feather coupling-entity. Not all non-manifold cases can be managed with this simplified version: two cells joined at a vertex are not allowed. An extra cycle pointer and the dual are necessary for valid navigation in a model. Another difference and a big advantage over the coupling-entity is that DHE (the full version) is able to represent cell/volume with a single dual vertex, and a face with a bundle (of edges). We also demonstrate (Boguslawski et al. 2010b) that our full system (with the dual) can represent certain 3D degeneracies in the model. Adding a ‘bridging edge’ in each face permits the construction of a hole through a shell (e.g. a torus) and adding a ‘bridging face’ permits the modelling of a completely enclosed cavity. (Adding these elements means we have a single connected graph, and may navigate, query and edit it.)

5 Construction of Models with Euler Operators Using standard Euler operators we can easily build polyhedra of any shape. Figure 5 shows the sequence used to create a cube from individual edge elements (this is one of many possible sequences). For clarity the dual is not shown, but it is present at each step, as with the external cell. The dual connects the internal and external cells together into one cell complex. Faces are defined automatically upon closure of the edges, and volumes are determined whenever a closed set of faces is completed. Holes and cavities (shown in Fig. 6) are allowed: we use a so-called ‘bridge edge’ (eb in Fig. 6a, b) to connect internal rings (holes) to the outside ring (the face); and a ‘bridge face’ (fb in Fig. 6b) to connect an internal cell (a cavity) with the outside cell. These permit us to represent many real-world situations.

52


Fig. 5 One of many possible ways to construct a cube using Euler operators

Fig. 6 Holes and cavities: (a) hole through polyhedra needs two faces with holes connected by bridge edges eb; (b) the front face with a hole – bridge edge eb connects internal loop lINT with the outside loop lOUT; (c) cavity in a cell – bridge face fb connects internal cell cINT with the outside cell cOUT


53

Fig. 7 Operators for a cell complex construction: (a) join/separate; (b) merge/split

To construct non-manifold models like two cells joined by a shared face an extension of the standard Euler operators set is necessary. Masuda (1993) presented a spanning set of extended Euler operators that allows for construction of complexbased non-manifold models. This set preserves topological consistency of the models. We use this idea to develop Euler operators to build a cell complex: we can join/separate (Fig. 7a) or merge/split (Fig. 7b) cells of the complex. Joining is an operation that glues two separate cells together. As result there are two connected cells with a double-sided face in between. Merging removes this shared face and the union of the two cells forms one volume. We perform a sequence of simple, atomic operations, the same as with standard Euler operators. ‘Join’ removes two adjacent faces from external cells; and combines two cells. Internal primal cells stay unchanged. ‘Merge’ removes the adjacent faces of two internal cells to form a single cell. Thus ‘Join’ operates on external cells while ‘Merge’ operates on internal ones. Lee (1999) shows a similar example using standard Euler operators (with no dual) for joining or separating cells.

6 Construction of Complex Building Interiors Our full model is capable of being used for rapid 3D building modelling, with full navigation and analysis. We take as our example two linked buildings on the University of Glamorgan campus. In total there are over 1,300 cells. (Cells may be rooms, doors or corridors.) The model, and its use for escape route planning, was prepared within 2 weeks. First the floor plans were scanned, and then manually traced (vectorised) using AutoCad, and then the doors were added. Then the rooms were extruded for each floor, within AutoCad. This produced a set of individual

54


Fig. 8 A building model with paths between two locations: (a) two rooms on the same corridor; (b) escape route from a room on the top floor; (c) connectivity via stairway


55

cells, one for each room – but not connected together. The various floors were adjusted to fit on top of each other, and exported as OBJ files to our software. Geometric intersection testing routines were used to check for adjacency. The results were used to construct the building models using our standard Euler operators, with both the primal and the dual graph being updated simultaneously. Because of our interest in escape route planning, as also in the case of Lee (2007), all doors on the plans were added as ‘flat’ (zero volume) cells, complete with their own centroids, to allow specification of navigation routes and to allow attributes, such as locking times, to be added. Corridors were broken into manageable sections by inserting ‘virtual doors’ appropriately. The resulting models functioned as desired. Figure 8 shows some views of the model with the shortest path visualized as grey cells (rooms, corridors, doors) along the path. These paths can be calculated ‘from room to room’ (Fig. 8a) – useful in building management for finding paths between two locations, or ‘from a room to an assembly point’ – [an escape route (Fig. 8b)]. Navigation between floors is via stairways (Fig. 8c); a stairway on one floor is connected to stairways below and above with horizontal ‘virtual doors’. As our original objective was escape route planning, we needed to describe the local terrain close to the buildings – e.g. for assembly points. This was achieved by adding thin cells (perhaps concrete paving) to the model, allowing navigation outside the building (Fig. 9). Applying Dijkstra’s algorithm to the dual graph allows the real-time calculation of the shortest path from any location to an assembly point.

Fig. 9 A building with an external terrain. (The exterior is composed of thin 3D cells, allowing navigation through them in the same way as with rooms)

56


7 Conclusions We have demonstrated that our new data structure, the dual half-edge, is flexible enough to permit rapid building model construction using Euler operators. Our illustrative application is escape route planning, where real-time navigation may be performed, the results changing as new accessibility information arrives during the emergency. The ability to apply attribute information to any model element (node, edge, face or volume – in either primal or dual space) opens a wide range of applications beyond disaster management. Examples include sound and radio wave propagation, virtual museums (with pictures/textures on each wall face), shopping centre assistance (with shop-face textures applied) and many more. Acknowledgements The first author’s research was supported by the Ordnance Survey and an EPSRC New CASE award.

References Boguslawski, P. and C. Gold, Construction Operators for Modelling 3D Objects and Dual Navigation Structures, in 3D Geo-Information Sciences. Lectures Notes in Geoinformation and Cartography. 2009, Springer, Berlin. pp. 47–59. Boguslawski, P. and C. Gold, Euler Operators and Navigation of Multi-shell Building Models, in Developments in 3D Geo-Information Sciences. Lecture Notes in Geoinformation and Cartography. 2010, Springer, Berlin. pp. 1–16. Boguslawski, P., C.M. Gold, and H. Ledoux, Modelling and analysing 3D buildings with a primal/ dual data structure. ISPRS Journal of Photogrammetry and Remote Sensing, 2010. Accepted (Scale, quality, and analysis aspects of city models), Article in Press. Guibas, L. and J. Stolfi, Primitives for the manipulation of general subdivisions and the computation of Voronoi. ACM Transactions on Graphics, 1985. 4(2): pp. 74–123. Kolbe, T.H., Representing and Exchanging 3D City Models with CityGML, in 3D Geo-Information Sciences, J. Lee and S. Zlatanova, Editors. 2009, Springer, Berlin, Heidelberg, pp. 15–31. Ledoux, H. and C.M. Gold, Simultaneous storage of primal and dual three-dimensional subdivisions. Computers, Environment and Urban Systems, 2007. 31(4): pp. 393–408. Lee, J., A three-dimensional navigable data model to support emergency response in microspatial built-environments. Annals of the Association of American Geographers, 2007. 97(3): pp. 512–529. Lee, K., Principles of CAD/CAM/CAE Systems. 1999, Addison-Wesley Longman, Reading. p. 582. M€antyl€a, M., Introduction to Solid Modeling. 1988, Computer Science Press, Rockville. p. 401. Masuda, H., Topological operators and Boolean operations for complex-based nonmanifold geometric models. Computer-Aided Design, 1993. 25(2): pp. 119–129. Yamaguchi, Y. and F. Kimura, Nonmanifold topology based on coupling entities. IEEE Computer Graphics and Applications, 1995. 15(1): pp. 42–50. Yamaguchi, Y., K. Kobayashi, and F. Kimura, Geometric Modeling with Generalized Topology and Geometry for Product Engineering, in Product Modeling for Computer-Aided Design and Manufacturing, J. Peger and J. Turner, Editors. 1991, Elsevier Science Publishers B.V., North-Holland.

Large Scale Constraint Delaunay Triangulation for Virtual Globe Rendering M. Christen and S. Nebiker

Abstract A technique to create a Delaunay triangulation for terrain visualization on a virtual globe is presented. This method can be used to process large scale elevation datasets with billions of points by using little RAM during data processing. All data is being transformed to a global spatial reference system. If grid based elevation data is used as input, a reduced TIN can be calculated. Furthermore, a level of detail approach for large scale out-of-core spherical terrain rendering for virtual globes is presented using the previously created TIN.

1 Introduction Terrain Rendering with very large and dense height data sets have become very popular due to the progression of geospatial imaging sensors such as airborne LiDAR (Fowler et al. 1997; Shan and Toth 2009). Terrain point densities from such sensors can be in the range of one or several points per square meter leading to massive data sets when applied to entire states or countries. Even for small countries, such a LiDAR data set could easily comprise of several hundred billions of points or several TB of data respectively. Despite the wide-spread use of TINbased elevation models in the GIS community, there are hardly any software solutions capable of efficiently processing such large data sets on the one hand of and also supporting the generation of levels of detail on the other. However, both capabilities are required to exploit such large TIN-based data sets in scalable and interactive 3D geoinformation environments. Virtual globes have a high impact on geographical 3D information systems and reality based games. They were mostly based on grid-based elevation data but will increasingly incorporate high-density elevation data sets in the multi-TB range. This paper shows an approach of how

M. Christen (*) and S. Nebiker Institute of Geomatics Engineering, University of Applied Sciences Northwestern Switzerland, Gr€undenstr. 40, 4132 Muttenz, Switzerland e-mail: [email protected], [email protected]


57

58

M. Christen and S. Nebiker

very large irregulary spaced high-density elevation datasets can efficiently be preprocessed for ellipsoidal rendering on a virtual globe.

2 Previous Work A virtual globe has been implemented at the University of Applied Sciences Northwestern Switzerland. This virtual globe is called i3D. Virtual globes can be implemented (Nebiker et al. 2007) using grid based geometry for the terrain. However, using regular grids has several drawbacks. For one the grid is limited to contain points aligned to the grid only: break-lines and spot heights are not supported since they are generally not grid-aligned. Another drawback is that when creating a polygonal representation from a grid representation many coplanar polygonneighbors may result, especially in flat areas like lakes, leading to expensive redundancy and unneeded geometry, which leads to wasted memory transfer between the CPU and GPU. Furthermore, if grid points are transformed from a local to a global spatial reference system the alignment of the grid changes to a new grid layout which requires height interpolations and thus leads to quality loss.

3 Related Work Numerous interactive terrain rendering methods have been proposed. A thorough survey on different approaches was recently made by Pajarola and Gobbetti (2007). Basically there are hierarchical level of detail approches using regular grids and others using irregular data sets (Hoppe et al. 1998; Pajarola et al. 2002). Lately there was more focus on GPU optimized techniques (Livny et al. 2009) and related topics like data compression between the CPU and GPU (Dick et al. 2009; Lindstrom et al. 2010). Most proposed works assume a planar reference – restricted to flat earth terrain rendering. Only few literature is considering spherical or even ellipsoidal planetary rendering or data processing, for example references (Gerstner et al. 1999; Szalay et al. 2007; Zhou et al. 2008). For data preprocessing, a streaming Delaunay computation of very large height data sets was introduced by Isenburg et al. (2006). Their approach calculates Delaunay triangulations by using the spatial coherence of the dataset itself – without previously sorting the data. However, their approach does not consider storing the resulting triangulation in a spatial data structure suitable for the virtual globe and applying further processing steps to the resulting TIN, like thinning out and calculating level of detail. Furthermore, an alternative way for visualizing large amount of LIDAR data in Google Earth has been presented by Isenburg et al. (2009). Their approach creates rasterized contour lines to visualize the data and not the actual geometry.

Large Scale Constraint Delaunay Triangulation for Virtual Globe Rendering

59

The Google Earth API (Google: Google Earth API Developer’s Guide) currently doesn’t have a mechanism to stream custom elevation data.

4 Virtual Globes 4.1

Overview of Virtual Globes

Virtual globes, sometimes also referred to as 3D geobrowsers, consist of virtual 3D environments capable of streaming and interactively displaying large amounts of geo-referenced spatial contents over the Internet (Nebiker et al. 2010). With the availability of the commercial products Google Earth (Google: Google Earth API Developer’s Guide) and Microsoft: Bing Maps 3d virtual globes gained enourmously in popularity. A virtual globe stores its data on servers that can be accessed by a client software. This software can either be a stand-alone executable file (Google Earth), or running from a browser plugin (Bing Maps, Google Earth). There is little reliable information available on how these commercial products create their large scale elevation models. This may be one of the reasons, why adding large custom-built elevation models is not possible with these globes.

4.2

Preprocessing Data for Virtual Globes

A virtual globe can have several data categories like image data, elevation data, points of interest, vector data, and 3D objects. Before streaming over the Internet this data must be preprocessed. This preprocessing usually comprises a transformation from a local to a global reference system, creation of pyramid layers or level of detail, tiling of the data, and optionally compression and encryption of the data.

4.3

Virtual Globe Tile Systems

Bing Maps, Google Earth, and i3D tiles are indexed using quadtree keys (quadkeys). Each quadkey number identifies a single tile at a single zoom level (see Fig. 1). Typically, the Mercator projection is used to map image and elevation data to a square area. The Mercator projection is mainly used to minimize distortions in processed images and elevation data. The meridians of the Mercator projection are vertical parallel equally spaced lines, cut at right angles by horizontal straight parallels which are increasingly spaced toward each pole so that conformality is preserved (Snyder 1987). The maximum latitude is chosen so that the resulting map fits into a square. For the spherical Mercator projection this maximum latitude is

60


Fig. 1 Bing maps tile system (Microsoft: Bing Maps Tile System)

approximately 85.05 and for the ellipsoidal Mercator projection it is approximately 85.08 . The projection is normalized to values in the range (1,1) to (1,1) to ensure increased numerical stability during the triangulation and thinning process. Bing Maps 3D uses the spherical Mercator projection and i3D supports both the spherical and ellipsoidal Mercator projection. Such an ellipsoidal geodetic reference model is required, in order to minimise geometric transformation errors and to enable position accuracies within the Virtual Globe at the sub-meter level. Listing 1 shows the implementation used in i3D. Points are transformed from the Mercator projection to WGS84 by using the respective algorithm described in Snyder (1987). 1 void EllipsoidToMercator (const double lngRad, const double latRad, double& out_x, double& out_y, const double e) 2{ 3 const double a ¼ 1.0 / M_PI; 4 const double lngRad0 ¼ 0; 5 6 if (e ¼¼ 0.0) // spherical case 7{ 8 out_x ¼ a* (lngRad-lngRad0); 9 out_y ¼ log(tan(M_PI/4.0 + latRad/2)); 10 } 11 else // ellipsoidal case, first eccentricity not 0. 12 { 13 out_x ¼ a*(lngRad-lngRad0);


61

14 out_y ¼ a*log (tan(M_PI/4.0+latRad/2.0)*pow((1.0-e* sin( latRad))/(1.0 + e* sin (latRad)),0.5*e)); 15 } 16 }

Listing 1 Transformation to normalized Mercator coordinates

5 Creating Large TIN for Virtual Globes In the design of elevation support for a Virtual Globe a number of requirements must be taken into account, many of which are distinctly different from those applicable to DEM support in standard GIS environments. These DEM requirements in Virtual Globes include: the support for highly accurate regional to global elevation models with a preservation of the original precision at a global scale; the possibility to add spot heights and breaklines – either during preprocessing or at runtime; the adaptive data reduction or data thinning in order to optimise storage space, transmission performance and memory utilisation in the rendering process; the support for level of detail and view-dependent multiple resolutions; the need for highly efficient processing of very large terrain models supporting parallelisation approaches; a streamable delivery and rendering of terrain data, and finally, the capability of merging and displaying multiple DEMs, e.g., a global model and a high-resolution local model, at run-time. In this section we document an approach addressing these specific requirements.

5.1

Choosing a Data Structure for Large Scale Triangulation

There are many possible data structures for representing triangulations. Thus, a triangulation data structure should be chosen in view of the needs and requirements of the actual application (Hjelle and Daehlen 2006). Popular data structures for triangulations are: triangle-based data structure with neighbors, vertex-based data structure with neighbors, half-edge data structure (Weiler 1985), and the quad-edge data structure (Guibas et al. 1983). For our large scale implementation of the Delaunay triangulation a trianglebased data structure is used. One of the reasons for that is that the triangle neighbor must support a mechanism to temporarily remove neighbor triangles from memory to simplify our large scale triangulation and to calculate level of detail which is easier to implement in a triangle based structure. The main characteristic of a triangle-based data structure is that edges are not stored directly – only vertices and triangles which are always stored in counterclockwise orientation as we can see in Listings 2 and 3.

62


The triangle structure is defined by using two structures, one for the vertex (also called elevation point) and a second structure for the triangle. 1 struct Vertex 2{ 3 double x, y; // position in mercator projection 4 double elv; // elevation above WGS84 ellipsoid [m] 5 double weight; // weight/importance of this point 6 }; Listing 2 Definition of an elevation point

All values of the vertex structure are held as double precision floating point to support precision in the millimeter range. Besides the position parameters a weight parameter is used which can be used to define the importance of the point. This is useful in the preservation of characteristic mountain peaks when calculating the level of detail. 1 struct Triangle 2{ 3 Vertex* pVertex0 ; 4 Vertex* pVertex1 ; 5 Vertex* pVertex2 ; 6 Triangle* pTriangle0 ; 7 Triangle* pTriangle1 ; 8 Triangle* pTriangle2 ; 9};

Listing 3 Definition of a triangle

The triangle structure holds pointers to vertices and neighbor triangles as shown in Fig. 2. Because vertex-pointers and triangle-pointers are shared among different triangles a reference count mechanism is implemented. Because the triangulation must be large scale, it is necessary to remove triangles in a special way by deleting them from memory without affecting the Delaunay condition. This can be done by deleting unused triangles from memory and by setting the appropriate triangle neighbor pointers to 0. This allows creating disconnected areas with islands in the triangle structure as shown in Fig. 3. Furthermore for thinning out data a mechanism must be available for removing points from the triangulation, for example by using the removal algorithm described in Mostafavi et al. (2003). The Delaunay triangulation itself is created using an incremental approach. In summary, the supported basic operations of the triangulation must be:


63

Fig. 2 Triangle data structure

Fig. 3 Disconnected area with island

l l l

Insert a new point into triangulation (for incremental construction) Remove a triangle from memory (for large scale support) Remove a point from triangulation (for thinning out data)

5.2

Incremental Delaunay Construction

Incremental Delaunay trigulation has been presented by Guibas and Stolfi (1985). For reasons of clarity it is described briefly below.

64


Fig. 4 Adding a new vertex to triangulation

The construction of an incremental Delaunay triangulation consists of adding new vertices to an already existing triangulation. When adding a new vertex to the triangulation, there are some cases to be distinguished (Fig. 4): 1. The new vertex is inside an existing triangle. In this case three new triangles are created. 2. The new vertex is on an edge of an existing triangle. In this case the triangles containing this edge are removed and four new triangles are created. 3. The new vertex has the same position as an existing vertex in the triangulation. In this case the new vertex is rejected. 4. The new vertex is outside a previously defined boundary. In this case the vertex is rejected. To maintain the Delaunay condition when inserting new points, invalid edges must be flipped, so that the circumcircle of a triangle never contains another point of the triangulation (Fig. 5).

5.3

Numerical Stability in a Global Spatial Reference System

Because the triangulation is done in a global spatial reference system, numerical stability is an important issue, as triangles may become very small in relation to the global system. When inserting a new vertex into the triangulation, cases 2 and 3 of the incremental Delaunay construction described above bear numerical limitations. If all points of the triangles are almost collinear (“skinny triangle”), numerical problems will occur when using standard floating point arithmetic. Exact computation of the incircle and orientation predicates makes the algorithm robust, but slow. More speed can be gained by using arithmetic filtering approaches (Shewchuk 1996; Devillers et al. 2003) (Fig. 5).

5.4

Large Scale Triangulation

Our algorithm is divided into three passes: 1. Geodetic transformation 2. Spatial subdivision 3. Triangulation and cell split


65

Fig. 5 Flipping edges to satisfy the Delaunay condition

First Pass: The first pass transforms points from the source spatial reference system to a normalized Mercator projection using any custom rotational ellipsoid or sphere as described in Listing 1. During this pass a new file containing all transformed points in Mercator projection is created. In addition, the axis aligned bounding box (in Mercator projection) of the dataset is calculated. In summary, at the end of the first pass, the total number of points and the axis aligned bounding box is known, and all transformed points are stored in a binary file. Second Pass: The second pass creates a spatial subdivision of the points. This spatial subdivision is the same global quadtree of the Mercator projection which is later used for out-of-core rendering of the terrain. All points are stored in tiles of the lowest level of detail of this dataset. The maximum number of points per tile can be specified and according to that number the lowest level of detail is calculated. At the end of the second pass the original dataset is spatially subdivided and for each tile a file exists. The filename of the tile is generated using its quad-code – a unique identifier for each cell in the quadtree (Fig. 6). Third Pass: The third pass actually calculates the triangulation. By traversing the quadtree structure in a Morton order(Morton 1966), the triangulation is calculated step by step. When creating tiles, new points are added to the triangulation, the border points of the tiles (see Fig. 7). These points are stored in an edge file. “Edge points” are usually shared by two neighbor tiles and “corner points” are shared by up to four neighbor tiles. Every tile has at least four corner points. If a corner is not part of the triangulation, a point with elevation 0 will be added and marked as “no data value”, otherwise points are calculated by doing a linear interpolation of the elevation points at the corresponding position in the triangle containing the point. These points are treated as a constraint in the Delaunay triangulation, so that a rectangular tile border exists for all tiles. Tiles are stored by saving corner points, divided into north edge points, east edge points, south edge points, west edge points, and interior points separately. This way the Delaunay triangulation can be recreated for every tile and therefore additional points or break-lines may be added at a later time or even interactively during visualization.

66


Fig. 6 Calculating triangulation: output of processing three levels. (Only white triangles of max. four tiles are kept in memory)

Fig. 7 Calculating “tile edges” and “tile corner points”

5.5

Creating Level of Detail Tiles

Once all tiles are stored, the level of detail for all remaining pyramid layers can be calculated. The maximum number of points per tile must be maintained through all levels of detail. The algorithm used is an adapted version of the mesh simplification method presented by Heller (1990). For each level of detail tile all points of the four parent tiles are loaded and merged together in a new tile. The interior of the new tile and the corner and edge points of the parent tiles are disregarded. The new tile is being triangulated and thinned out by using the algorithm described by Lee (1989): for each point in the new triangulation an error value is calculated. This error value


67

holds the elevation difference which results when the point is removed from the triangulation. Points with small error are removed first. This procedure is repeated until the number of points in the tile is smaller than or equal to the maximum allowed number of points per tile. If the error value is still smaller than a previously defined epsilon value the points may even be thinned out more, which allows thinning out flat areas like lakes in an early stage of the level of detail.

6 Rendering the Virtual Globe Because visualization may consist of terabytes of orthophoto and elevation data, out-of-core rendering with a level of detail approach must be used. Data can be streamed over the Internet, a local network or a local hard drive. This is done by using download threads as shown in Fig. 8. Using a least recently used caching approach (LRU), the most recently requested data remains in the most expensive and fastest memory layer. For the level of detail approach a quadtree is used. The quadtree can be mapped to the WGS84 ellipsoid and an error metric adapted to ellipsoids decides which resolution is suitable for the current view. Depending on the error per pixel, a new resolution is requested from another data storage layer. The data itself is stored in tiles based on a quadtree. Tiles with neighbors of a different level of detail must be stitched together to avoid visualization artifacts. This usually involves updating the geometry of the tile with higher resolution to match the neighbor resolution, or using a curtaining method to hide stitching problems in a more efficient manner as it is not necessary to figure out the resolution of neighbors and to avoid retriangulations during visualisation.

Fig. 8 Elevation tile data is being downloaded while the rendering thread remains running at constant speed

68


7 Results The presented approach for large scale constraint Delauney triangulation has been implemented in our own i3D virtual globe technology.

7.1

Streaming Data

The propoposed algorithm produces tiled data that can be streamed from a harddisk, local area network, or from the Internet. Depending on the number of cores of the client machine, up to eight download threads are created to retrieve the data while the main thread is responsible for rendering data available in best resolution as shown in Fig. 8.

7.2

Quality

The error metric allows to change a quality parameter in real-time. This can be used to force level of detail to a specific setting, as shown in Fig. 9.

Fig. 9 Interactive change of quality parameters. Using swisstopo DHM25 dataset. Base data # swisstopo (JA100071)


69

Computers or mobile devices with lower memory or low-end graphics hardware can be set to a lower quality while maintaining a certain quality. Lower quality is also preferred in a slower network as it results in faster streaming.

7.3

Combining Image Data

The produced elevation tiles can be combined with image data, the result is shown in Fig. 10. Image data is being tiled using the same quadtree tiling structure as the elevation data. Image and elevation tiles are not stored together to be independent. It is possible to exchange image tiles without updating elevation tiles. In Fig. 11 elevation tiles and image tiles are blended together to see the resolution difference between image and elevation data.

8 Conclusion An approach for creating large scale elevation data processing has been introduced. With this approach it is possible to process global and local elevation datasets for visualization on a virtual globe. It permits the efficient processing of very large scale regularly and irregularly spaced terrain data sets on standard computer

Fig. 10 Rendering a TIN (original DEM point spacing of 25 m) on the virtual globe with and areal imagery with 50 cm per pixel resolution. Using SWISSIMAGE and DHM25 datasets. Base data # swisstopo (JA100071)

70


Fig. 11 TIN wireframe and image data blended together. Using SWISSIMAGE and DHM25 datasets. Base data # swisstopo (JA100071)

hardware. The presented triangulation approach is well suited for a future parallelisation of the triangulation of national to continental high-resolution elevation data sets. The presented TIN-based approach offers a number of advantages over gridbased terrain rendering in earlier versions of virtual globes, namely a significantly improved terrain representation with a simultaneous dramatic reduction in data size and streaming performance. The approach also supports the on-the-fly integration of spot heights and break-lines at runtime. This on-the-fly integration of breaklines, including special geospatial features such as Terrain Intersection Curves (TIS) of the CityGML standard (2010), will be a prerequisite for enabling future applications in Virtual Globes requiring more accurate and higher fidelity representations of urban environments (Nebiker et al. 2010).

References Devillers, O., Devillers, Pion, S., Pion, S., Prisme, P.: Efficient exact geometric predicates for Delaunay triangulations. In: Proceedings of the 5th Workshop Algorithm Engineering and Experiments. pp. 37–44. Baltimore (2003) Dick, C., Schneider, J., Westermann, R.: Efficient geometry compression for GPU- based decoding in realtime terrain rendering. Comp. Graph. Forum 28(1), 67–83 (2009)


71

Fowler, R.E., Samberg, A., Flood, M., Greaves, T.J.: Modeling mobile terrestrial LiDAR to vector based models. In: Maune, D. F. (ed.) Digital Elevation Model Technologies and Applications: The DEM Users Manual, chap. Topographic and Terrestrial Lidar. pp. 199–252. American Society of Photogrammetry and Remote Sensing, Bethesda (1997) Gerstner, T.: Multiresolution visualization and compression of global topographic data. Tech. rep., GeoInformatica (1999) Google: Earth, http://earth.google.com Google: Google Earth API Developer’s Guide, http://code.google.com/apis/earth/documentation/ Guibas, L.J., Stolfi, J.: Primitives for the manipulation of general subdivisions and the computation of voronoi diagrams. In: STOC’ 83: Proceedings of the Fifteenth Annual ACM Symposium on Theory of Computing. pp. 221–234. ACM, New York (1983) Guibas, L.J., Stolfi, J.: Primitives for the manipulation of general subdivisions and the computation of Voronoi diagrams. ACM Trans. Graph. 4, 74–123 (1985) Heller, M.: Triangulation algorithms for adaptive terrain modeling. In: Proceedings of the 4th International Symposium on Spatial Data Handling. pp. 163–174. Zurich, Switzerland (1990) Hjelle, O., Daehlen, M.: Triangulations and Applications (Mathematics and Visualization). Springer-Verlag New York, Secaucus, NJ (2006) Hoppe, H.: Smooth view-dependent level-of-detail control and its application to terrain rendering. In: VIS ’98: Proceedings of the Conference on Visualization ’98. pp. 35–42. IEEE Computer Society Press, Los Alamitos (1998) Isenburg, M., Liu, Y., Shewchuk, J., Snoeyink, J.: Streaming computation of Delaunay triangulations. ACM Trans. Graph. 25(3), 1049–1056 (2006) Isenburg, M., Shewchuk, J.: Visualizing LIDAR in Google Earth. In: Proceedings of the 17th International Conference on Geoinformatics. Fairfax (2009) Lee, J.: A drop heuristic conversion method for extracting irregular networks for digital elevation models. In: Proceedings of the GIS/LIS ’89. pp. 30–39. Orlando (1989) Lindstrom, P., Cohen, J.D.: On-the-fly decompression and rendering of multiresolution terrain. In: I3D ’10: Proceedings of the 2010 ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games. pp. 65–73. ACM, New York (2010) Livny, Y., Kogan, Z., El-Sana, J.: Seamless patches for GPU-based terrain rendering. Vis. Comput. 25(3), 197–208 (2009) Microsoft: Bing Maps 3d, http://www.bing.com/maps Microsoft: Bing Maps Tile System, http://msdn.microsoft.com/en-us/library/bb259689.aspx Morton, G.: A computer oriented geodetic data base and a new technique in file sequencing. Tech. Rep. IBM Ltd., Ottawa, Ontario, Canada (1966) Mostafavi, M.A., Gold, C., Dakowicz, M.: Delete and insert operations in Voronoi/Delaunay methods and applications. Comput. Geosci. 29(4), 523–530 (2003) Nebiker, S., Christen, M., Eugster, H., Fl€ uckiger, K., Stierli, C.: Integrating mobile geo sensors into collaborative virtual globes – design and implementation issues. Paper presented at the Mobile Mapping Technologies Symposium MMT 2007, Padua (2007) Nebiker, S., Bleisch, S., Christen, M.: Rich point clouds in virtual globes a new paradigm in city modeling? Computers, Environment and Urban Systems (June 2010), http://dx.doi.org/ 10.1016/j.compenvurbsys.2010.05.002 Open Geospatial Consortium, Inc: OpenGIS® city geography markup language (CityGML) – encoding standard (ogc 08-007r1). (p. 218): Open Geospatial Consortium Inc. (2010) Pajarola, R., Gobbetti, E.: Survey of semi-regular multiresolution models for interactive terrain rendering. Vis. Comput. 23(8), 583–605 (2007) Pajarola, R., Antonijuan, M., Lario, R.: Quadtin: quadtree based triangulated irregular networks. In: VIS ’02: Proceedings of the Conference on Visualization ’02. pp. 395–402. IEEE Computer Society, Washington, DC (2002) Shan, J., Toth, C.: Topographic laser ranging and scanning. CRC Press, Boca Raton (2009) Shewchuk, J.R.: Adaptive precision floating-point arithmetic and fast robust geometric predicates. Discrete Comput. Geometry 18, 305–363 (1996)

72


Snyder, J.P.: Map Projections: A Working Manual. U.S. Geological Survey Professional Paper 1395, U.S. Geological Survey, http://pubs.er.usgs.gov/usgspubs/pp/pp1395 (1987) Szalay, A.S., Gray, J., Fekete, G., Kunszt, P.Z., Kukol, P., Thakar, A.: Indexing the sphere with the hierarchical triangular mesh. CoRR abs/cs/0701164 (2007) Weiler, K.: Edge-based data structures for solid modeling in curved-surface environments. IEEE Comput. Graph. Appl. 5(1), 21–40 (1985) Zhou, Q., Lees, B., Tang, G.A.: Lecture Notes in Geoinformation and Cartography, chap. A Seamless and Adaptive LOD Model of the Global Terrain Based on the QTM. pp. 85–103. Springer Berlin Heidelberg, New York (2008)

Towards Interoperating CityGML and IFC Building Models: A Unified Model Based Approach ¨ stman, and Khurram Shahzad Mohamed El-Mekawy, Anders O

Abstract CityGML represents 3D urban objects that can be shared over different applications, whereas, IFC provides a very detailed semantic model for 3D building representations using constructive elements like beams, walls, etc. Attempts have been made to interoperate CityGML and IFC for seeking useful common applications. However, these efforts use a unidirectional method (mostly from IFC to CityGML) for conversion processes. A bidirectional method can lead to development of unified applications in the areas of urban planning, building construction analysis, homeland security, etc. The benefits of these applications clearly appear at the operational level (e.g., cost reduction, unified data-view), and at the strategic level (e.g., crisis management and increasing the analyses capabilities). In this paper, we present an approach for interoperating CityGML and IFC based on development of a unified building model for converting IFC to CityGML and vice versa. The conversion is a two-steps process in which a model is firstly converted to the unified model and secondly to the target model. Finally, we demonstrate the approach and outcome of each step by a hospital building case that is located in Norrt€alje City, north of Stockholm, Sweden.

M. El-Mekawy (*) Future Position X, G€avle, Sweden e-mail: [email protected] ¨ stman A. O GIS Institute, University of G€avle, G€avle, Sweden e-mail: [email protected] K. Shahzad Department of Computer and System Sciences, Royal Institute of Technology (KTH), Stockholm, Sweden e-mail: [email protected]


73

74

M. El-Mekawy et al.

1 Introduction By merging building models and geographic information models, public and private enterprises are attempting to fulfil the application demands in construction analysis, urban planning, homeland security, etc. (Benner et al. 2005; Isikdag and Zlatanova 2009a). Geometric models have been developed in both domains but semantic models are relatively few. The two most prominent semantic models are Industry Foundation Classes (IFC) and City Geography Markup Language (CityGML) (Isikdag and Zlatanova 2009a). The goal of IFC is to specify a common language for building industry technology that improves communication, productivity, delivery time, cost, and quality throughout the design, construction and maintenance life cycle of buildings (Hallberg and Tarandi 2009). IFC is then used to assemble computer readable models that contain related information to different parts of a building (Karola et al. 2002; IFC Model, http://www.iai-tech.org/ifc/IFC2x3/TC1/html/, accessed 01-2011). CityGML is however used to define information related to topological and semantic properties of a geographical area including buildings (OGC, http:// www.opengeospatial.org/; Gr€ oger et al. 2008). On one hand, IFC has been developed as an ISO standard and it has been largely accepted for the building industry (Isikdag and Zlatanova 2009a). On the other hand, CityGML has been recently adopted as an international standard for modelling cities in the Open Geospatial Consortium (OGC) and the EU (OGC, http://www.opengeospatial.org/; Gr€oger et al. 2008). This has led to major increase in the development of new application areas, software tool kits and extensions to existing systems for supporting IFC as well as CityGML. The integration of IFC and CityGML is seen today as a needed step for getting a more complete picture of 3D modelling at different levels of detail, i.e., sharing and exchanging information between building industry objects (represented in IFC) and geospatial object (represented in CityGML). Several efforts have been made to integrate CityGML and IFC. All these efforts are mainly in form of developing frameworks, extended discussion for addressing requirements, or developing conversion tools. For frameworks, the IFC for GIS (IFG) project was initiated by The Norwegian Strate Planning Authority (Statens Bygningstekniske Etat) and completed in 2007. This framework aimed to exchange building information between CAD systems and GIS using IFC. The project succeeded to create a mapping specification from XML version of IFG geometry to GML and vice versa (IFG, http://www.iai.no/ifg/index_history.html). Looking specifically on technical aspects, another framework was proposed by Nagel (2007) and aimed for algorithms that automatically transform IFC building models into CityGML models (Nagel 2007). In 2009, Isikdag and Zlatanova did complement Nagel’s framework by proposing a framework for automatic generation of buildings in CityGML using BIM based on definition of building semantics and components (Isikdag and Zlatanova 2009a). Following the holistic view of 3D city modelling aspects, an

Towards Interoperating CityGML and IFC Building Models

75

extended discussion on conceptual requirements for converting CityGML to IFC models was proposed in 2009 by a team led by Thomas Kolbe at the Technical University of Berlin (Nagel et al. 2009). They proposed a framework that integrates 3D graphics/data of buildings and urban areas (X3D, DXF, KML, COLLADA, etc.) with semantic data in a CityGML target schema. Additional to that, a few conversion environments can be seen in this area. León (2009) demonstrated his team’s latest Application Domain Extensions (ADE) that integrates Building Information Model (BIM) data based on the open standard IFC into CityGML (Van Berlo 2009). Not only research efforts, but commercial software products for conversion from IFC to CityGML [e.g., IfcExplorer (http://www.ifcwiki.org/index.php/IfcExplorer_ CityGML_Export) and Safe Software (http://www.safe.com/products/desktop/ formats.php)] also contribute to the development of 3D city modelling integration. However, these attempts have either; (a) an approach for a unidirectional conversion with a focus on converting geometries (mostly from IFC to CityGML), (b) a discussion about what should be done in terms of integration, i.e., how it should be done is not sufficiently implemented yet, (c) focused on down-grading IFC to lower LoDs (LoDs) in CityGML, or (d) a discussion on the interest of rich semantics of IFC. Several studies (Benner et al. 2005; Isikdag and Zlatanova 2009a; Nagel et al. 2009) have emphasized that a formal framework for strict semantic and geometry conversion is required for a complete integration of CityGML and IFC. As a consequence, the purpose of this paper is to propose and describe a unified model oriented approach that can be used for bidirectional conversion between IFC and CityGML. The proposed approach thereby contributes towards increasing integration of CityGML and IFC for extending 3D city model applications. Geometry is highlighted as one of the main problematic concerns for integrating IFC and CityGML (Nagel et al. 2009; Isikdag and Zlatanova 2009b). However, most of the recent efforts have focused on the conceptual integration or conversion processes. Considering the vast amount of efforts for defining geometric differences and developing conversion algorithms (Lapierre and Cote 2008), this type of problem is not in our objectives. In this study we instead focus our discussion on the conceptual integration and mapping of different objects in both IFC and CityGML standards. The rest of the paper is organized as follows: Sect. 2 discusses how we approach the research problem. Building models for CityGML and IFC are presented in Sect. 3, followed by detailed discussion on UBM in Sect. 4. In Sect. 5 the two-steps approach for conversion of IFC to CityGML and CityGML to IFC is presented along with an illustrative case study.

2 Research Approach A unified model is here defined as a superset model concept that is extended to contain all the features and objects from both IFC and CityGML building models. It is an intermediate model relating objects from both models. The unified model

76

M. El-Mekawy et al.

Fig. 1 The research approach

is originally derived from the superset concepts of mathematics in 1970s led by Thomas J. Jech and other researchers (Jech 1971; Miguel et al. 2002). It has also been used in software engineering from the mid 1990s. Due to its wide acceptance, Unified Modelling Language (UML) has been selected as the modelling language for our Unified Building Model (UBM) approach. The requirements on the proposed UBM are that conversion is done with a minimum information loss and an efficient schema matching and mapping process. The approach consists of the following steps as shown in Fig. 1: (1) elicitation of IFC building model, (2) development of the UBM, (3) conversion between IFC building model and UBM, and (4) conversion between UBM and CityGML building model. Prior to discussing CityGML and IFC conversion, it is necessary to discuss each model separately.

3 CityGML and IFC Building Models In order to interoperate CityGML and IFC it is essential to develop building models for both. This section discusses the development of building models for IFC and CityGML.

3.1

IFC Building Model

IFC is an object oriented format developed by the International Alliance for Interoperability (IAI) (http://www.iai-tech.org/). It is used to facilitate interoperability in


77

the building industry and sharing information among participants. IFCs are used to assemble computer readable models that contain data elements that represent parts of buildings and their relevant information (Lapierre and Cote 2007). There is no universally accepted building model for IFC (Kolbe et al. 2008). In this section we, however, present an IFC building model that is primarily based on the work done by the IAI and ISO in form of IFC standard documentation (IAI, http://www.iai-tech.org/), the ISO 16739 standard (http://www.iso.org/iso/iso_catalogue/catalogue_tc/ catalogue_detail.htm?csnumber¼38056) and Benner et al. (2005). From the standards we identify important concepts (Fig. 2). UML standard notations are used for developing the IFC building model. A building should have at least one storey and may have multiple storeys. Each building storey may have zero or more spaces related to it, i.e., a building structure which has only one wall is a building with zero spaces. Building elements and opening elements are subtypes of structural element. Each building element has zero or more opening elements, i.e., a wall without any door or window has zero openings, whereas each opening element (like door, window) is attached to only one building element. Figure 2 shows 12 types of building elements that can represent a building structure in IFC standard. The scope of this study is limited to building models that only represent constructed parts of buildings.

3.2

CityGML Building Model

CityGML is an open standard that has been implemented as an application schema for GML3 (OGC, http://www.opengeospatial.org/; Gr€oger et al., Open Geospatial Consortium). GML3 is the extendible international standard for spatial data exchange that has been developed and issued by the Open Geospatial Consortium

Fig. 2 IFC building model

78

M. El-Mekawy et al.

CityObject

Site

MultiSurface 0..1

* Room

*

AbstractBuilding

Door

*

0..2

Window

*

0..1

InteriorWallSurface FloorSurface

*

* * Opening

0..1

CeilingSurface

* * BoundarySurface

* BuildingInstallation

RoofSurface WallSurface GroundSurface ClosureSurface

LoD-1

LoD-2

LoD-3

LoD-4

Fig. 3 CityGML building model

(OGC) (Cox et al. 2004) and the ISO TC211 (http://www.isotc211.org/). CityGML is developed as an open data model expressed by an XML schema. CityGML files can store and exchange virtual 3D objects and city models among applications (CityGML, http://www.citygml.org). Based on ISO 19107 (http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_ detail.htm?csnumber¼26012) and ISO 19109 (http://www.iso.org/iso/catalogue_ detail.htm?csnumber¼39891), a CityGML Building Model has been produced in the CityGML standard (Kolbe 2008). The building model (shown in Fig. 3) is an excerpted version from the CityGML standard in which only the used conversion concepts to IFC are represented, i.e., BuildingFurniture, and IntBuildingInstallation are not represented. UML standard notations are used for developing CityGML building model. CityGML is developed in five levels of detail (LoDs) which are used to represent city model objects according to required details’ level in different applications. LoDs are numbered from LoD0 to LoD4 and they have different accuracies and minimum dimensions of spatial objects that can be represented in each LoD. Individual buildings start to appear at LoD1. Therefore, we have presented CityGML building model from LoD1 to LoD4. Figure 3 represents the LoDs by different colours.

4 The Unified Building Model IFC holds more detailed information about building objects than CityGML. Therefore, IFC to CityGML conversion is a less complex task as compared to CityGML


79

to IFC conversion. Due to that, most of previous work has aimed at IFC to CityGML conversion. However, two-ways conversion approach between CityGML and IFC has not been sufficiently explored. In this section, we present the proposed Unified Building Model (UBM) and justify the inclusion of its elements. The UBM is capable of capturing information about spatial structures and building objects from both IFC and CityGML building models. The model is, therefore, used as an intermediate step for conversion of IFC to CityGML and vice versa. UBM can be used as a starting point to support applications where information from both views (CityGML and IFC) is required for analyses. It can also facilitate modelling a database schema that is capable of capturing information that is required for all level of details. In the absence of UBM, if the application is designed at a specific LoD (e.g., LoD2) the information about higher level of detail (e.g., LoD4) cannot be acquired. To build the UBM, all classes with their concepts were initially collected from both models while omitting their relationships. Overlapping concepts were merged and new objects were created in such a way that both indoor and outdoor objects are captured. Finally, relationships between the objects were redefined to produce our UBM. UML notations are used for representing its objects and relationships between them. Figure 4 shows the UBM where different colours are used to represent different LoDs. The UBM is briefly described below. As the study is limited to building structure, the objects beyond building (such as project and site) are not represented in the model. As building is a common object for both IFC and CityGML, UBMbuilding object has been used as a starting point for the model. A building in CityGML consists of rooms, where a room is a space surrounded by different boundary surfaces. Storeys are not explicitly defined but they can be represented as an explicit aggregation of all building features on a

Fig. 4 The unified building model

80

M. El-Mekawy et al.

certain height level (Gr€ oger et al. 2008). However, in IFC, the structure is smoothly organized by breaking down a building into storeys and then into spaces that form a specific storey. In the UBM, we consider concepts from IFC as well as CityGML. A building in the UBM consists of explicit definition of storeys, at least one storey in a building. Each building storey may have zero or more spaces related to it. A space can be either opened or closed as shown in Fig. 4. The closed space represents a room which corresponds to the room definition in CityGML. In CityGML, boundary surfaces are used by the class (_BoundarySurface) for structuring the exterior shell of a building and visible surfaces of a room. However, in IFC, a room or space is built by building elements that structure and surround it. We can then highlight that some of the IFC classes (e.g., IfcWall, IfcDoor and IfcWindow) are (or handled as) boundary surfaces in CityGML. In the UBM we have used concepts from both CityGML and IFC. We propose that a boundary surface is only used when it is needed to extract the box model of a complete building, part of a building or a specific storey. Building elements, however, are used to represent the elements which exist in the building structure. As every spatial place in a building (storey or space) may have a door or a window, opening is connected to both boundary surface and building element objects. For building elements, we have defined some new concepts in such a way that all concepts from both IFC and CityGML can be covered. The main difference between building elements in CityGML and IFC is the representation of different surfaces, interior and exterior parts of a building (wall, roof and ground). Because of the need for different LoD representations and definitions of elements like (roof, ceiling, and slab) and (ground, floor and slab), we have defined the building elements as follows: l

l

l

Covering is a closing level that covers a space from the top side. It has two types; Roof for the top covering of a building or the top storey which gives the external shape of a building from above, and Ceiling for the covering of any space in a building. In both subtypes only constructed parts for a space are considered in the covering representation. Level is a walkable (not only horizontal) level that represents bottom level of a space. It has two types; Ground for the bottom level of ground floor which has a connection to the outer ground to give the external shape of a building from bottom level, and Floor for the bottom level of a space in any space of a building except the bottom (lowest) storey. Wall is a vertical/semi-vertical element that surrounds or subdivides spaces. It has three subtypes; (a) CurtainWall for the outer wall that covers a complete facade of a building or a part of it. It is usually not limited or attached to only one storey, (b) Interior for an internal wall between rooms or spaces (none of its faces has connection with the outer environment), and (c) Exterior for an external wall that has connection with the outer environment and represents a part of external facades of a building.

Towards Interoperating CityGML and IFC Building Models UBMWindow

UBMBuilding

UBMDoor

81 UBMCovering

UBMRoof

UBMLevel

UBMCeiling

1 UBMBoundary Surface

1..* UBMStorey 0

*

UBMGround

*

* *

UBMWall

1

1..* *

0..* UBMSpace

0..2

UBMOpening

UBMBuilding Element

*

*

UBMCurtain Wall

UBMBuildingInstallation

*

UBMOpened Space

UBMFloor

UBMInterior Wall UBMClosed Space

UBMStair

UBMRamp

UBMBeam

LoD-1

LoD-2

UBMRailing

UBMColumn

LoD-3

UBMExterior Wall

LoD-4

Fig. 5 Covering, level and wall concepts

The above used concepts are shown in Fig. 5. Building installations (ramps, chimneys, balconies, beams, column, etc.) are defined differently in both IFC and CityGML. In IFC, they are defined as normal building elements (like walls, slabs, etc.) with the same geometric concepts. However, in CityGML building installations are specified in a separated object named building installation. In the latter case, geometries of these installations (ramps, chimneys, balconies, beams, column, etc.) are defined by multi surfaces that construct the objects and stored in CityGMLMultiSurface. In the UBM, we have defined these installations as subclasses of UBMBuildingInstallation class in order to represent the external building installations that are important for the external shape of a building at LoD2. This also allows the turning-off of building installations if a smaller model is desired for faster processing or less detailed analysis. When merging classes, constraints among attributes may be defined. Attributes of the class UBMBuilding is for instance based on the merging of IFCBuilding and CityGMLAbstractBuilding. The following constraints may then be defined: UBMBuilding:Envelope¼IfcBuilding:Envelope¼CitygmlAbstractBuilding:Envelope UBMBuilding:Id¼½IfcBuilding:Id;CitygmlAbstractBuilding:Id

82

M. El-Mekawy et al.

5 Conversion Between IFC and CityGML In this section, we present our method for converting IFC to CityGML and for converting CityGML to IFC. IFC to CityGML conversion is done in two steps, i.e., IFC model is converted to the UBM, and the UBM is then converted to CityGML. Similarly, the CityGML to IFC conversion is done in two steps. We demonstrate the use of our approach by a case study of a hospital building. In the two following subsections, we present the proposed steps with their conversion procedures. Locum Company (Locum AB), an owner of hospitals’ buildings in Sweden, decided to adopt IFC standard in all of their upcoming new public buildings (http:// www.locum.se). Modelling of existing buildings is still under discussion as it requires much more effort and updates of different data sources. In a research initiative, the first hospital building to model on IFC was the Norrt€alje City Hospital in the north of Stockholm. A full IFC model for the hospital building was built (Fig. 6).

5.1

Conversion from IFC to CityGML

Step1. As a first step of IFC to CityGML conversion, we define a set of rules which describe how the objects in the UBM can be produced from IFC model. At LoD1 of the UBM, a building is represented by the UBMBuilding object as a box/block model with a solid geometry. The needed information for constructing the box model can be acquired from attributes of different objects in IFC. For example, from IFCWall, IFCSlab and IFCRoof objects, it is possible to extract the construction information and coordinates of external walls and building roof. At this LoD, each building element (walls and roofs) should be simplified in its IFC

Fig. 6 IFC model of Norrt€alje hospital


83

model into a single flat surface. These surfaces are used as surrounding surfaces of the box model. As what is needed at this LoD is only the outer envelope of a building, extension objects such as balconies and chimneys and geometrical curvature should be eliminated from the model. Therefore, the bounding boxes of the IfcWall elements combined with the top and ground floors from IFCSlab entity can formulate the outer envelope which is used for the box model. Alternatively, IfcRoof can also be used to form the boundary representation (BRep) box of the whole building by simplifying the roof shape, extracting its height and projecting it on the ground of the building. This may be written as: f : O ½IfcWalls; IfcRoof ! O UBMBuilding where O x represents the domain of x At LoD2 of the UBM, external objects and their details are represented, i.e., curves in roofs, surfaces of grounds and external details of a wall. The needed information for UBMBoundarySurface can be acquired from a combination of the building elements like IFCRoof, IFCWall and IFCSlab. The needed information for UBMBuildingElements can be acquired from building elements of IFC. However, at this level only exterior elements (see Fig. 4) are presented. For example, UBMExteriorWall and UBMCurtainWall should be represented whereas UBMInteriorWall should not. Information about different parts of the same wall that form a facade can be obtained from combined/Union-of IfcWall objects. These parts can be aggregated and stored in UBMBoundarySurface because they can be used as boundary surfaces to different rooms and spaces in the building. Additional to that, information about every part, which is obtained from IfcBuildingElement, will be referenced using UBMBuildingElement for the purpose of using them as separated building elements. IfcCurtainWall can be stored in UBMCurtainWall because they might be used as bounding surfaces with specific dimensions and attributes. Similarly, IfcRoof and IfcSlab can be aggregated and simplified to form UBMCovering.Roof and UBMLevel.Ground respectively. Exterior building installations at this LoD represent an important part of the building image. Information about UBM building installations can be obtained from IfcStair, IfcRamp, IfcRailing, IfcBeam and IfcColumn. To differentiate between exterior and interior building installation, we however propose a generation of two projection footprints, vertical and horizontal. The vertical footprint is a result of projecting vertical structure elements, e.g., IfcWalls and IfcColumn, whereas, the horizontal footprint is a result of projecting horizontal structure elements, e.g., IfcSlab and IfcBeam. Using geometrical information, building elements are then checked versus the two projected footprints. Only connected objects will be represented as exterior building installations and others are omitted. Information about external objects and building installations with their surfaces will be then stored in UBMBuildingElements and UBMBoundarySurface. Rules may be formally written as:

84

M. El-Mekawy et al.

fAggregation : OIfcWalls ! OUBMBoundarySurface f : OIfcBuildingElement ! OUBMBuildingElement fCombination : O½IfcWalls; IfcRoof; IfcSlab ! OUBMBoundarySurface f f f f

: : : :

OIfcWall ! OUBMExterior OIfcCurtainWall ! OUBMCurtainWall OIfcSlab ! OUBMLevel:Ground OIfcRoof ! OUBMLCovering:Roof

f : O½IfcStair; IfcRamp; IfcRailing; IfcBeam; IfcColumn ! O UBMBuildingInstallation½UBMStair; UBMRamp; UBMRailing; UBMBeam; UBMColumn

At LoD3 of the UBM, parts of external objects (that remains unrepresented at LoD2) are represented. Internal elements of building structure and opened spaces are represented as well. The information needed for UBMStorey can directly be acquired from IfcBuildingStorey. The information needed for UBMSpace.Opened can be acquired from a combination of IfcSpace and IfcOpeningElement. In IFC, opening elements (IfcOpeningElement) are attached to building elements. Inside each opening element, there is usually one filling element (IfcWindow or IfcDoor). IfcOpeningElement is used itself as an element to describe the geometry and semantics of the opening. In this case, IfcOpeningElement may contain multiple IfcDoor and IfcWindow elements with referencing their geometry. Two classes in IFC define the representation of doors and windows. The first, IfcRelVoidsElement, is a one-to-one linking relationship between an element that contain an opening element (e.g., a wall) and one opening element that creates a void in the element. The second, IfcRelFillsElement, is another one-to-one relationship between an opening element and an element that fills that opening (Fig. 7). Geometric information about doors and windows are defined as Sweeping and CSG models as other building elements (e.g., walls and slabs). In CityGML, Window and Door objects are defined as subclasses of the abstract class Opening. They are, however, represented by different surfaces and modelled by gml::MultiSurface RelatedElements

IfcRelContained InSpatialStructure

RelatedElements

IfcWallStandardCase

IfcDoor / IfcWindow

RelatingBuildingElement

RelatingBuildingElement

IfcRelVoidsElement RelatedOpeningElement

IfcRelFillsElement RelatedOpeningElement

IfcOpeningElement Element Voiding

Fig. 7 Opening concepts in IFC

Element Filling


85

geometry. In our proposed UBM, we apply the same concept for opening elements as in CityGML. Therefore, the Door and Window objects in the UBM can be generated by acquiring information from classes of IfcDoor and IfcWindow. However, conversions from Sweeping/CSG geometric models that are used in IFC to BRep models that are used in CityGML have to be performed. Formally, the conversion rules can be written as follows: f : OIfcBuildingStorey ! OUBMStorey fCombination : O½IfcSpace; IfcOpeningElement ! OUBMSpace:Opened f : OIfcBuildingElement:IfcDoor ! OUBMOpening:UBMDoor f : OIfcBuildingElement:IfcWindow ! OUBMOpening:UBMWindow At LoD4 of the UBM, all objects of a building structure, i.e., the interior walls, floors, ceilings, etc. are represented. In IFC, there is no concept of specific Room object as in CityGML. IfcSpace class defines all volumes and areas that are bounded actually or theoretically. This definition includes rooms that are bounded by different building elements. As all IFC objects, Sweeping/CSG geometry is used for spaces and their elements which requires conversions to BRep geometric models that we use in the UBM. For differentiating between exterior and interior elements, the same concept of generating two projected footprints is used as at LoD2. The vertical footprint results projecting vertical structure elements, whereas, the horizontal footprint results from projecting horizontal structure elements. The geometries of these two footprints are checked to extract all building elements that intersect with them. Only connected objects will be stored as exterior building elements (i.e., walls and roof). To create spaces in the UBM, information from IfcWall, IfcRoof and IfcSlab that form the boundaries of rooms are used. Information about both UBMCeiling and UBMFloor can be acquired from IfcSlab where a slab may represent both a ceiling for a storey and a floor for another storey on the top of it (for example, a ceiling for the second storey is a floor for the third storey). Geometries of elements that form each IfcSpace are checked. If they close a shape (coordinates of starting point is the same as of ending point), the space is stored in UBMSpace.Closed class. Otherwise, it is stored in UBMSpace.Opened class. It is important here to mention that information of all building elements that form a space is stored in UBMBuildingElement class and all boundaries of the room are represented within the classes aggregated under the UBMBoundarySurface class. They are connected through the UBMSpace class. It is worthwhile to mention here that for simplification, BuildingFurniture and IntBuildingInstallation are not included in our model. Formally, the transformation rules may be written as: fCombination : O½IfcSpace; IfcWall; IfcRoof; IfcSlab ! OUBMSpace:Closed f : OIfcWall ! OUBMWall:Interior f : OIfcSlab ! OUBMCovering:Ceiling f : OIfcSlab ! OUBMLevel:Floor

86

M. El-Mekawy et al.

The first step in this process shows conversion rules for converting IFC building model to different LoDs in the UBM. To demonstrate the use of step1, we have applied the transformation rules on the hospital building case previously discussed. Figure 8 shows a set of building elements conform to the proposed UBM for the Norrt€alje Hospital Building at LoD4. Step2. As a second step of IFC to CityGML conversion we define a set of rules which describe how the objects in the CityGML model can be produced from UBM. Due to space limitations we provide only the set of rules and limited discussion about important rules in the form of a table. Table 1 shows the transformation rules for step2 of converting IFC to CityGML. On applying these transformation rules, Fig. 9a–d shows the hospital building in CityGML at LoD1 to LoD4 respectively. Figure 9e shows an interior view of the building at LoD4 which represents the interior faces of building elements.

Fig. 8 A part from Norrt€alje hospital showing UBM objects

Table 1 Transformation rules from UBM to CityGML LoD Rule LoD1 f : OUBMBuilding ! OCityGMLAbstractBuilding LoD2 f : OUBMBoundarySurface ! OCityGMLBoundarySurface f : OUBMCovering.Roof ! OCityGMLRoofSurface f : OUBMWall.Exterior ! OCityGMLWallSurface f : OUBMLevel.Ground ! OCityGMLGroundSurface f : OUBMBuildingInstallation ! OCityGMLBuildingInstallation LoD3 f : OUBMOpening ! OCityGMLOpening f : OUBMDoor ! OCityGMLDoor f : OUBMWindow ! OCityGMLWindow LoD4 F: OUBMSpace.Closed ! OCityGMLRoom f : O[UBMBoundarySurface, UBM.BuildingElement] ! OCityGMLMultisurface f : OUBMCovering.Ceiling ! OCityGMLCeilingSurface f : OUBMWall.Interior ! OCityGMLInteriorWall f : OUBMWall.Floor ! OCityGMLFloorSurface


87

Fig. 9 (a) LoD1-box for the Hospital building. (b) LoD2 for the hospital building. (c) LoD3 for the hospital building. (d) LoD4 for the hospital building – external view. (e) LoD4 for the hospital building – internal view

88

5.2

M. El-Mekawy et al.

Conversion from CityGML to IFC

Similar to the conversion from IFC to CityGML, a two-steps method is also carried out to convert CityGML to IFC. In the first step, UBM is produced from CityGML and then, IFC model is produced from UBM in the second step. Due to space limitations we provide only a set of rules with limited discussion for the important ones in the form of a table. One can ask here why we need to go from the a very generic level (i.e., LoD1 and LoD2 of CityGML) to a more detailed model (i.e., IFC). However, we argue that these conversions result in models which are represented in their exterior structures. In Architecture, Engineering and Construction (AEC) applications, such generic models can be useful for analysis of testing the prototyping of building structures, templates for architectural interior designs, 3D interior logistics and utility planning, and hence, supporting the unified applications that integrate IFC and CityGML in city modelling applications. We, therefore, start by describing the conversion from CityGML LoD1 and LoD2 to the proposed UBM. Step1. As a first step of CityGML to IFC conversion, we define a set of rules that describe how the objects in the UBM can partially be produced from CityGML model. The rules are given in the following description. At LoD1, as mentioned in (Sect. 5.1), a building is presented by a box/blocks model. Information needed for constructing the box model should be stored in the UBMBuilding object. This object conforms to the CityGML AbstractBuilding object. Different structural entities of exterior surfaces of a building are aggregated to simple boxes without any details. As an alternative, information of the exterior shell of a building can be stored in CityGMLMultiSurface. In this case, the information can be transformed to UBMBuilding object attributes. Conversion rules may be written as: f : OCityGMLAbstractBuilding ! OUBMBuilding At LoD2, the exterior shell of a building starts to be decomposed into details. CityGMLBoundarySurface represents the base class for all other objects that form a building shell. At this LoD, the only needed details are about; the covering of a building and shapes of its roof (from CityGMLRoofSurface), exterior walls with their details (from CityGMLWallSurface) and the ground floor shape of the building (from CityGMLGroundSurface). By applying conversion of BRep to Sweeping/ CSG geometric models, information of these building elements are transformed to the UBMBuildingElement class and their surfaces are stored in the UBMBoundarySurface. For building installations, the concept in the UBM is very close to that of CityGML. Therefore, we can use one-to-one mapping of the CityGMLBuildingInstallations geometries into UBMBuildingInstalltion geometries. In some cases, building installations are not needed at this LoD, so they can then be omitted from the model. When only external installations are needed, similar two-footprints


89

process that is discussed in (Sect. 5.1 LoD2 and LoD4) can be used. Only connected objects with these two footprints will be stored as exterior building installations. However, we suggest that in case of small models, all building installations can be added at LoD2. Formally, conversion rules may be written as: f : OCityGMLRoofSurface ! OUBMCovering:Roof f : OCityGMLGroundSurface ! OUBMLevel:Ceiling At LoD3, CityGMLBoundarySurface holds information about different bounding objects of different spaces. Information about objects that share the same horizontal level in one building storey (or may be based on user definitions) can be aggregated to form UBMStorey. In IFC a storey is defined in three types; complex (when it spans overall several connected storeys), element, (when all building storeys are similar in their structure and height) and partial (when it is decomposed in horizontal parts). Considering these types and their definitions, the UBMBuildingStorey is geometrically described by all spaces that share the same floor surface. These spaces will be referenced from their boundary surfaces of CityGMLRoofSurface, CityGMLWallSurface and CityGMLGroundSurface which define the external shell of each storey. In LoD4, however, these objects will be supported by the internal objects for more detailed floors. Opened spaces can be also formed at this level in a condition of a room in CityGML that does not have six bounding objects (four walls or sides of walls, ceiling and a floor) or does not have a closing geometry of its bounding elements. Alternatively, classification of opened and closed spaces can be done manually or to be left for LoD4 as division of open spaces (e.g., dividing a long corridor in different open spaces). At this LoD, opening elements should also appear in the model. Our concept of opening elements is similar to that of CityGML as door and window are subclasses of the class opening. Therefore, one-to-one mapping can be done between CityGMLOpening, CityGMLWindow and CityGMLDoor into corresponding classes in the UBM. Formally, conversion rules may be written as: fAggregationWithConditions : OCityGMLBoundarySurface ! OUBMStorey fAggregation : O½CityGMLBoundarySurface; CityGMLBuildingElements ! OUBMSpace:Opened f : OCityGMLOpening:Door ! OUBMOpening:Door f : OCityGMLOpening:Window ! OUBMOpening:Window LoD4 represents the highest detailed LoD in which all the building elements, exteriors and interiors, should be represented. We have solved the problem of splitting an IFC element (for example a wall) in its exterior and interior faces by adding, in the UBM, exterior and interior classes as subtypes of wall. An opposite process should be carried out to aggregate different faces of a building element (for example wall) to form and IFC element. We, however, propose the generation of two projected footprints as it is introduced above in (Sect. 5.1 LoD2 and LoD4) to

90

M. El-Mekawy et al.

differentiate between exterior and interior elements. Information about building elements themselves is stored in different subclasses of the UBMBuildingElements class. In this case conversion from BRep to Sweeping/CSG geometric models are required. However, their surfaces information will be also referenced in the UBMBoundarySurface. For the current step, to convert from CityGML to the UBM, we do not show transformation of all building elements, as objects in CityGML can be matched easily to their corresponding UBM objects, for example CityGMLCeiling is matched to UBMCovering.Ceiling. Rooms at this level should be converted to closed spaces in the UBM. We have discussed differentiation between opened and closed (room) spaces at LoD3. Formally, rules may be written as: f : OCityGMLCeilingSurface ! OUBMCovering:Ceiling f : OCityGMLInteriorWallSurface ! OUBMWall:Interior f : OCityGMLFloorSurface ! OUBMLevel:Floor f : OCityGMLRoom ! OUBMSpace:Closed Step2. As a second step of CityGML to IFC conversion we define a set of rules which describe how the objects in the IFC model can be produced from the proposed model. Due to space limitations we provide only the set of rules and limited discussion about important rules in the form of a table (Table 2).

6 Discussion and Conclusion IFC and CityGML represent indoor and outdoor spatial objects of a building. In order to fulfil the demands in urban planning applications and construction analysis, it is important to integrate IFC and CityGML. However, existing approaches do not provide complete integration because they mostly offer a unidirectional transformation, i.e., from IFC to CityGML. There is a significant overlapping for information content for mapping IFC and CityGML to the UBM. However, there is no one-to-one mapping for all data. While there are concepts in the UBM are adopted as in IFC, there are others that are closer or adopted from CityGML definitions. The first step for this mapping is to extract Table 2 Transformation rules from UBM to IFC

LoD LoD3

LoD4

Rule f : OUBMOpening ! OIfcOpening f : OUBMOpening.Door ! OIfcDoor f : OUBMOpening.Window ! OIfcWindow f : OUBMStorey ! OIfcBuildingStorey f : OUBMSpace ! OIfcSpace f : OUBMCovering.Ceiling ! OIfcSpace f : OUBMCovering.Roof ! OIfcCovering f : OUBMFloor ! OIfcSlab f : OUBMWall.CurtainWall ! OIfcCurtainWall f : OUBMWall.Interior ! OIfcWall


91

the relevant semantic objects from IFC and CityGML that contain needed information and their geometries. That is followed by identifying the level of detail (LoD) which we target in the UBM. Table 3 presents the corresponding elements for the mapping from IFC and CityGML into the UBM. In this paper we have presented an approach based on a Unified Building Model (UBM) for reaching interoperability between IFC and CityGML. It is a two-steps method that can be used for converting IFC to CityGML building models and vice versa. By using this approach a building model is first converted to the UBM, followed by conversion to the target building model. Limitations of our approach are, (a) CityGML to IFC is not fully demonstrated, (b) UBM should also be tested on several buildings, (c) the conversion process is not verified. The identified limitations are also our future research directions. From the study, we can conclude the following: (a) the approach is a candidate for bidirectional conversion of CityGML and IFC, and (b) the approach provides a starting point towards complete integration of CityGML and IFC. The latter Table 3 IFC – UBM – CityGML mapping IFC UBM IfcBuilding UBMBuilding IfcBuildingStorey UBMStorey

IfcSpace

IfcSlab (Ground Slab) (Floor Slab) (Ceiling Slab) IfcRoof IfcWall (Exterior Wall) (Interior Wall) IfcCurtainWall IfcOpeningElement IfcDoor IfcWindow IfcBeam IfcColumn IfcCovering IfcStair IfcRailing IfcRamp

UBMSpace UBMOpenedSpace UBMClosedSpace UBMLevel UBMGround UBMFloor UBMCovering UBMCeiling UBMCovering UBMRoof UBMWall UBMExteriorWall UBMInteriorWall UBMWall UBMCurtainWall UBMOpening IfcDoor IfcWindow UBMBuildingInstallation UBMBuildingInstallation UBMBuildingInstallation UBMBuildingInstallation UBMBuildingInstallation UBMBuildingInstallation

CityGML _AbstractBuilding BoundarySurface RoofSurface WallSurface GroundSurface Other building elements

Room GroundSurface FloorSurface CeilingSurface RoofSurface

WallSurface InteriorWallSurface WallSurface Opening Door Window BuildingInstallation BuildingInstallation BuildingInstallation BuildingInstallation BuildingInstallation BuildingInstallation

92

M. El-Mekawy et al.

conclusion represents how different classes, attributes and relations have been considered from IFC and CityGML when building the UBM. Additional constraints are also added in such a way that may lead, with future development, to full semantic integration of the building models. To illustrate the applicability of our proposed approach, we have used a hospital case study. The purpose of the case study is to present how different building elements at different levels of detail can be rebuilt. Considering future research possibilities, we look to our integration approach as a starting point for developing a common database that formulates a UBM’s platform. With such platform, data from IFC and CityGML can be automatically integrated and processed in different analyses. Other formats can also be included in further steps. We finally believe that our approach needs future research efforts beyond only the building models and on implementation process for testing and verification. Acknowledgement This paper has been prepared as a part of research work in the Future Position X project NYSTA 39686, financially supported by the European Union structural funds, objective 2.

References Benner, J., Geiger, A., Leinemann, K.: Flexible generation of semantic 3D building models. In: Gr€oger, G. et al. (Eds.): Proceedings of the 1st International Workshop on Next Generation 3D City Models, Bonn (2005). CityGML homepage, http://www.citygml.org [last accessed 01-2010]. Cox, S., Daisy, P., Lake, R., Portele, C., Whiteside, A.: OpenGIS GeographyMarkup Language (GML3), Implementation Specification Ver 3.1.0, OGC Doc.No. 03-105r1. 2004. Gr€oger, G., Kolbe, T.H., Czerwinski, A., Nagel, C.: OpenGIS City Ge-ography Markup Language (CityGML) Encoding Standard, Version 1.0.0, OGC Doc. No. 08-007r1, Open Geospatial Consortium (2008). Hallberg, D., Tarandi V.: On the use of 4D BIM in LMS for Construction Works, Journal of Information Technology in Construction (Itcon), 2009 IAI, http://www.iai-tech.org/ [last accessed 01-2010]. IFC Model documentation webpage. International alliance for operability, 2008, http://www. iai-international.org/Model/R2x3_final/index.htm [last accessed 01-2010]. IFCExplorer CityGML Export, http://www.ifcwiki.org/index.php/IfcExplorer_CityGML_Export [last accessed 01-2010]. IFG, http://www.iai.no/ifg/index_history.html [last accessed 01-2010]. Isikdag, U., Zlatanova, S.: Towards defining a framework for automatic generation of buildings in CityGML using BIM. In: Lee, J., Zlatanova, S. (Eds.): 3D Geo-information Sciences, LNG&C, Springer-Verlag, pp. 79–96 (2009a). Isikdag, U., Zlatanova, S.: A SWOT analysis on the implementation of BIM within geospatial environment. In: Krek, A., Rumor, M., Zlatanova, S., Fendel, E. (Eds.): Urban and Regional Data Management, UDMS Annuals 2009, CRC Press, Boca Raton, FL, pp. 15–30 (2009b). ISO 16739 Industry Foundation Classes, Release 2x, Platform specification, http://www.iso.org/iso/ iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber¼38056 [last accessed 11-2009]. ISO 19107 Geographic information – spatial schema, http://www.iso.org/iso/iso_catalogue/ catalogue_tc/catalogue_detail.htm?csnumber¼26012 [last accessed 11-2009]. ISO 19109:2005 Geographic information – rules for application schema, http://www.iso.org/iso/ catalogue_detail.htm?csnumber¼39891 [last accessed 11-2009].


93

ISO TC211, Geographic information/geomatics, http://www.isotc211.org/ [last accessed 11-2009]. Jech, T.: Lectures in set theory, Springer-Verlag Lecture Notes in Mathematics 217 (ISBN 9783540055648) (1971). Karola, A., Lahtela, H., H€anninena, R., Hitchcockb, R., Chenc, Q., Dajkadand, S., Hagstr€ome, K.,: BSPro COM-server-interoperability between software tools using industrial foundation classes. Energy and Buildings 34(9), 901–907 (2002). Kolbe, T. H.: Representing and Exchanging 3D City Models with CityGML, In: Lee J., Zlatanova, S., Lecture Notes in Geoinformation and Cartography, 3D Geo-Information Sciences, Springer, Berlin-Heidelberg (2008) Kolbe, T.H., Nagel, C., Stadler, A.: CityGML – A framework for the representation of 3D city models from geometry acquisition to full semantic qualification. In: Proceedings of ISPRS Congress 2008, Bejing, China (2008). Lapierre, A., Cote, P.: Using open web services for urban data management: a testbed resulting from an OGC initiative for offering standard. CAD/GIS/BIM services. In: Coors, V., Rumor, M., Fendel, E.M., Zlatanova S. (Eds.): Urban and Regional Data Management. Proceedings of the 26th UDMS, October 10–12, Stuttgart (2007). Lapierre, A., Cote, P.: Using open web services for urban data management: a testbed resulting from an OGC initiative offering standard CAD/GIS/BIM services. In: Coors, V., Rumor, M., Fendel, E., Zlatanova, S. (Eds.): Urban and Regional Data Management, Taylor & Francis, London, pp. 381–393 (2008). Locum, A.B., Sweden’s larger property managers, http://www.locum.se [last accessed 01-2010]. Miguel, M., Jourdan, J., Salicki, S.: Practical Experiences in the Application of MDA, Fifth International Conference on the Unified Modeling Language – the Language and Its Applications, Dresden, Germany (2002). Nagel C.: Conversion of IFC to CityGML; Meeting of the OGC 3DIM Working Group at OGC TC/PC Meeting, Paris (Frankreich), Juli (2007). Nagel, C., Stadler, A., Kolbe, T.: Conceptual Requirements for the Automatic Reconstruction of Building Information Models from Uninterpreted 3D Models, Academic Track of Geoweb 2009 Conference, Vancouver (2009). OGC, http://www.opengeospatial.org/ [last accessed 01-2010]. Safe Software, FME Desktop Translator/Converter Software, http://www.safe.com/products/ desktop/formats.php [last accessed 01-2010]. Van Berlo, L.: CityGML extension for Building Information Modelling (BIM) and IFC. Free and Open Source Software for Geospatial (FOSS4G), Sydney (2009).

.

Initial Investigations for Modeling Interior Utilities Within 3D Geo Context: Transforming IFC-Interior Utility to CityGML/UtilityNetworkADE Ihab Hijazi, Manfred Ehlers, Sisi Zlatanova, Thomas Becker, and León van Berlo

Abstract 3D City models have so far neglected utility networks in built environments, both interior and exterior. Many urban applications, e.g. emergency response or maintenance operations, are looking for such an integration of interior and exterior utility. Interior utility is usually created and maintained using Building Information Model (BIM) systems, while exterior utility is stored, managed and analyzed using GIS. Researchers have suggested that the best approach for BIM/GIS integration is harmonized semantics, which allow formal mapping between the BIM and real world GIS. This paper provides preliminary ideas and directions for how to acquire information from BIM/Industry Foundation Class (IFC) and map it to CityGML utility network Application Domain Extension (ADE). The investigation points out that, in most cases, there is a direct one-to-one mapping between IFC schema and UtilityNetworkADE schema, and only in one case there is one-to-many mapping; related to logical connectivity since there is no exact concept to represent the case in UtilityNetworkADE. Many examples are shown of partial IFC files and their possible translation in order to be represented in UtilityNetworkADE classes.

I. Hijazi (*) and M. Ehlers Institute for Geoinformatics and Remote Sensing – IGF, University of Osnabrueck, Osnabrueck, Germany e-mail: [email protected], [email protected] S. Zlatanova OTB, Research Institute for Housing, Urban and Mobility Studies, Delft University of Technology, Delft, The Netherlands e-mail: [email protected] T. Becker Institute for Geodesy and Geoinformation Science, Technische Universit€at Berlin, Berlin, Germany e-mail: [email protected] L. van Berlo Netherlands organisation for applied scientific research (TNO), Delft, The Netherlands e-mail: [email protected]


95

96

I. Hijazi et al.

1 Introduction BIMs are becoming increasingly popular in the geospatial community, and have allowed for the production of a certain amount of work integrating BIM into GIS (Benner et al. 2005; Clemen and Gr€ undig 2006; Industry Foundation Classes for GIS (IFG) 2009; Isikdag et al. 2008; Isikdag and Zlatanova 2009; Peachavanish et al. 2006; Wu and Hsieh 2007). BIM is seen as an essential data source for creating a more navigable, interactive and visually realistic information for built environments (Peachavanish et al. 2006). Moreover, some building elements, such as interior utility networks, e.g. gas, water, and electricity, are hidden. The geometry of the existing network is not visible and cannot be collected using traditional 3D measurement and re-construction methods (i.e. laser scanning, surveying, photogrammetric techniques). Therefore, there has been an increasing interest in addressing the issues related to interoperability and integration between 3D BIM data and 3D GIS data. Integrating building utilities within their broader context will open the doors to new application thereof for management of urban utilities (e.g. maintenance operation, emergency response, inspection operation). Several real-world cases can highlight some specific issues, which require integration of interior and exterior utilities in a broader context (Hijazi and Ehlers 2009; Hijazi et al. 2009). City authorities perform regular inspections of some buildings in the city (e.g. chemical labs, factories) to ensure that the water discharged from these buildings is not wasting the public water resources. The inspection team needs to find the location of these elements inside these buildings (trace from specific point on exterior network to locate the equipment in building), in order to test whether they are working probably. Other case refers to maintenance operation in large campuses e.g. university, hospital. Facility managers need to perform maintenance operations, which can be either caused by a failure in the network or planned (preventive) operation (due to date of expiration or cleaning). Both cases will cause an outage of service; since replacing of elements is required. Therefore, buildings occupants need to be warned. Integrating interior and exterior utilities will allow performing analysis and defining the equipment that would be out of service to the room level (e.g. labs, offices). Currently, facility management make an assumption and generalize the announcement, and some time been unable to provide information to the concerned persons. It is widely known that overcoming the problems associated with heterogeneous environments requires interoperable methodologies and tools. Standardization of data models has been suggested and practiced as a major stride towards achieving the goal of interoperability (Akinci et al. 2008; Peachavanish et al. 2006; Shen et al. 2009). Industry Foundation Classes (IFCs) (Liebich et al. 2010) and City Geographic Markup Language (CityGML) (Gr€ oger et al. 2008) are officially recognized as two standards, and have been independently developed; the former by the International Alliance for Interoperability (IAI), which is the standardization body for the AEC/FM community, and the latter by the Open Geospatial Consortium (OGC), which is the standardization body for the geospatial community. As a result, IFCs are not readily usable in GIS and CityGML (Akinci et al. 2008; Isikdag and

Initial Investigations for Modeling Interior Utilities Within 3D Geo Context

97

Zlatanova 2009; Nagel et al. 2009). The main problem in the integration of BIMs with geospatial information occurs at the point of transfer of geometric information (Wu and Hsieh 2007). Building models use representations such as CSG and Sweep Geometry and in local coordinate system, while geospatial models mainly use Boundary Representation (BRep) and in real world coordinate systems. A utility network can be modeled in CityGML (Gr€oger et al. 2008) using UtilityNetworkADE (Becker et al. 2010) (First Draft) – a generic utility network model that is under development (ADE 2010). The IFC schema has different entities that can support the GIS utility network application. IFC schema contain entities representing different network objects, and classify them based on semantics, e.g. pipes, lamps, although connectivity concepts are also involved (Liebich et al. 2010). A thorough understanding of semantics is required to achieve integration and schema mapping (converting models, objects, or descriptions from one world into the concepts used in the other world). In this paper, we investigate the possibility of integrating the 3D BIM utility network data into a GIS. CityGML Utility Network Application domain extension (ADE) has been employed in this study. The development of the mapping follows a pragmatic approach by means of a manual inspection of both schemas to see which entities and attributes correspond. The work includes analyzing the different concepts of building service system presented in IFC, and looking for possibilities to represent these concepts in CityGML/UtilityNetworkADE. The paper is structured as follows: after the introduction, Sect. 2 provides an overview of related work with respect to BIM-GIS, CityGML and possible transformation between both standards. While Sect. 3 presents the main concepts and modeling paradigms of IFC concerning utilities, Sect. 4 introduces the main concepts of CityGML’s UtilityNetworkADE. In Sect. 5 we present the investigation results concerning the mapping from IFC utility to CityGML utility, it is organized in two subsections: thematic classes and graph structures. In Sect. 6 the research is concluded.

2 Related Work To date, a certain amount of work has been carried out on the integration of IFC, GIS and CityGML. Two approaches have been employed: either to transfer geo-data from GIS to BIM; or to transform BIM data into GIS. The industrial foundation class for GIS (IFG) (Industry Foundation Classes for GIS 2009) was one of the first efforts for the integration of IFC models with GIS (and geospatial models). Benner et al. (2005) proposed the QUASY object data model, which is a 3D semantic building model for urban development. The authors also developed a tool for automatic generation of semantic building models based on IFC-models. Nagel et al. (2009) demonstrate converting IFC to CityGML, and focus mainly on the CityGMLBuilding class. He provides a conceptual requirement for the automatic reconstruction of building information models from uninterrupted 3D Models. Their two-step

98

I. Hijazi et al.

transformation strategy incorporates CityGML as an intermediate layer between 3D graphic models and IFC/BIM models. Wu and Hsieh (2007) presented an algorithm for automatic conversion of IFC to GML. The algorithm works only on the transformation from swept solid models to the B-Rep. Isikdag and Zlatanova (2009) provide preliminary ideas for defining semantic mapping between the IFC data model and CityGML. The purpose of their work was to allow automatic transformation between the two models. In another study, Isikdag et al. (2008) have investigated the possibilities of automatic conversion from IFC to GIS based on two case studies: fire response and site selection. Their research includes a description of a developed tool for automatic conversion from IFC to ESRI shapefile and GeoDatabase. Another effort for integrating BIM in CityGML is GeoBIM ADE. Its purpose is to extend CityGML with the rich semantics of IFC (ADE 2010). In parallel, commercial software for conversion from IFC to CityGML and vice versa is in development [i.e. IFCExplorer (2008), BIMserver (2009), and FME from Safe Software (2008)]. However, the focus of the transformation methods presented in the above studies considers only the building’s architectural elements, such as walls, spaces, doors, also it concentrating on the geometry transformation issues. To our knowledge, there is no systematised study on enabling interoperability between IFC and GIS for utility networks.

3 Interior Building Utility in BIM/IFC An international standard for data exchange of building information model (BIM) data is the industrial foundation class (IFC). It was developed by the International Alliance of Interoperability (IAI) to facilitate interoperability in the building industry (Liebich et al. 2010). The goal of the IFC is to enable interoperability between building information systems. The latest release is IFC 2x3 TC1. Version 2x3 has introduced the ifcXML specification by using XML schema to define the IFC models in parallel with EXPRESS (Wu and Hsieh 2007). By contrast with the traditional de facto standards of CAD exchange, such as drawing files dxf or dgn, the IFC is strictly model-based. A wall is not a set of lines (polygons), but rather an object with specified attributes and relations (Clemen and Gr€undig 2006). The key contents of the current IFC 2x3 include (Isikdag et al. 2008; Liebich et al. 2010; Schevers et al. 2007; Shen et al. 2009): l l l l

l

IFC enables re-use of building information through the whole building lifecycle IFC is an object-oriented and semantically rich model IFC provides the representation of building models in 3D IFC is a spatially related data model, wherein spatial relationship between building elements are maintained in a hierarchical manner IFC provide a schema for electrical wiring and plumbing details

IFC supports utility networks inside buildings (building service systems); it has a shared building service element layer that defines the basic concepts required for


99

interoperability between building service domain extensions: i.e. the plumbing, Heating Ventilation Air Conditioning (HVAC), Electrical and Control domains. IFC methodology for specializing building service components follows a general approach for a component within a distribution system, no matter what the systems is. Therefore, the same IFC component can be shared and used to represent the different networks (e.g. gas, water and electricity) (see Fig. 1). The IFC concept for utility support has the following major characteristics (Liebich et al. 2010): l

l

l

All building service elements are exchanged by a subtype of IfcDistributionElement. The various subtypes inherit attributes and relationships from IfcDistributionElement and do not add any further attribute. These subtypes are introduced for the purpose of semantics and logical structuring of the model. Examples of these subtypes include IfcFlowSegement, which defines the occurrence of a segment of a flow distribution system that is typically straight, contiguous and has two ports (e.g. a section of pipe or duct). IfcFlowController defines the occurrence of elements of a distribution system that are used to regulate flow through a distribution system (e.g. damper, valve, switch, relay). Its type is defined by IfcFlowControllerType or its subtypes. IFC handle connectivity using two methods, logical and physical connectivity. Logical connectivity can be achieved without having a physical element to realize the connectivity. On the other hand, physical connectivity is achieved by enhancing the logical connectivity with a realizing element. IFC represents building objects by considering the real 3D shape; such as CSG, Sweeping, or SolidModel. One object can have several representations. Building IfcProperty

IfcObject IfcProject

IfcProductRepresentation

IfcEnergyConversionDevice

IfcBuildingStorey

IfcProduct

IfcLocalPlacement

IfcFlowTerminals

IfcSpace

IfcFlowFittings

IfcFlowController

IfcSpatialStructureElement IfcDistributionChamperElement

IfcFlowStorageDevice

IfcSystem IfcSpaceBoundary

IfcFlowTretmentDevice IfcElement

IfcFlowSegment

IfcFlowMovingDevice IfcOpeningElement IfcSlab

IfcWall

IfcBuildingElement

IfcDistribuationElement

IfcDistribuationFlowElement

Fig. 1 Informal IFC schema, the highlighted part is representing the hierarchy chart of interior utility – building service elements

100

l

I. Hijazi et al.

service components implement SolidModel representation using specific types, which are: B-Rep, Surface model and Sweep solid. Each of these types can be replaced by Mapped Representations, which allows the reuse of the shape of a particular object. The object can be inserted one or several times by using a block reference, including a transformation matrix. IFC provides the concept of system; a system is defined as an ‘organized combination of related parts within an AEC product, composed for a common purpose or function or to provide a service’.

4 Utilities Within CityGML–UtilityNetworkADE CityGML is an OGC standard, which provides a specification for the representation of 3D urban objects (Gr€ oger et al. 2008). It is the only 3D information model for the exchange of 3D city models. One of the reasons for creating such a model was to enrich 3D city models with thematic and semantic information. The information model of CityGML is an XML-based format implemented as an application schema of Geography Markup language (GML3). Today, CityGML seems to provide the best framework for semantic-geometric relations of 3D objects above earth surface (Emgard and Zlatanova 2008; Groneman and Zlatanova 2009). It maintains a good taxonomy and aggregations of Digital Terrain Models, sites (including buildings), vegetation, water bodies, transportation facilities, and city furniture. The underlying model differentiates five consecutive levels of detail (LOD), where objects become more detailed with increasing LOD regarding both geometry and thematic differentiation. In LODs 2–4 of CityGML the building facade is defined in the form of boundary surfaces, i.e. WallSurface, Roof Surface, Ground Surface or Closing surface. The LOD4 allows the representation of interior building elements, e.g. rooms, furniture, interior wall surfaces. Nevertheless, the current version of CityGML still lacks the integration of subsurface features, such as geology, utility networks and underground constructions (e.g. tunnels). Recently the modeling group of SIG 3D has been working on some extensions of CityGML; these include extensions to represent bridges (ADE 2010), and utility networks. The latter is the extension to model utility networks in cities (Becker et al. 2010). The intuitive efforts have resulted in the publishing of the first draft version for utilizing the CityGML concept of application domain extension to provide an abstract level data model that provides the main concepts to model utility networks regardless of its type. The ADE forms a new package, providing the possibility to integrate one or more Utility Networks into a city (Fig. 2). The data model has the following features (Becker et al. 2010). l

l

It provides a graph data structure with simple geometry representations, as well as true geometry representation of utility network objects. The network features, which represent any topographic object in the network (e.g. pipe, tunnel), are inherited from the cityobject class, just as are other classes such as streets or buildings.


101

Fig. 2 UtilityNetworkADE and its relation to CityGML and GML (Becker et al. 2010) l

l

All the utility network elements are generalized (core model), with no classification of elements based on their specialized functions in the network. The data model allows the aggregation of a utility network element to represent a specific network as well as the aggregation of many features to represent a superior feature

5 Mapping IFC Building Service Systems Data into UtilitiesNetworkADE–CityGML Based on the previous description of the features of IFC and CityGML standards regarding utility networks, there is a significant overlapping of information content between both standards. This section compares the different UtilityNetworkADE classes and the corresponding information in IFC. We investigate the possibilities for generating the UtilityNetworkADE in CityGML using the IFC utility network classes (see Fig. 3). The mapping described on this paper is grouped into two categories thematic classes and graph structure.

102

I. Hijazi et al. core::_CityObject

gml::_Feature

> NodeType +exterior +interior

gml::_FeatureCollection

> InterFeatureLinkType 0..*

_NetworkFeature

1

1 consistsOf

0..* 1 0..1

topoGraph

topoGraph 1

realization

1

0..1

start

1

1

0..*

1

end 0..*

_Edge

realization 0..1

_LinkControl

+type: NodeType +connectionSignature: _Signature[0..1] +linkControl: _LinkControl [0..1]

gml::Point gml::_Curve

0..1

1

Node

nodeMember

_signature

NetworkGraph

featureGraphMember 1..*

1

1

0..*

subNetwork

0..*

FeatureGraph

Network

*

component

+connects +contains

+direction : gml::SignType [0..1] +linkControl: _LinkControl [0..1]

linkMember

linkMember 0..* 0..*

InteriorFeatureLink

NetworkLink

+type : InterFeatureLinkType

linkMember

Constraints: Both nodes must belong to the same FeatureGraph

InterFeatureLink

0..*

Constraints: 1. Each node must belong to a different FeatureGraph 2. Each node type must be exterior 3. The connectionSignature of both nodes must be compatible / identical 4. Both nodes must belong to the same Network

Constraints: 1. Each node must belong to a different FeatureGraph 2. Each node type must be exterior 3. The connectionSignature of both nodes must be compatible / identical 4. Each node must belong to a different Network

Fig. 3 UML class diagram for the UtilityNetworkADE the upper part represent the topographic classes and the lower part represent the graph representation (Becker et al. 2010)

5.1

Thematic Classes

UtilityNetworkADE of CityGML uses two classes to represent networks; these are _NetworkFeature and Network. The class Network is a central element in the UtilityNetworkADE data model. It is intended to provide a topographic representation of entire networks (e.g. water, gas). It is derived from the GML Feature Collection, and serves therefore as a collection of networks, where each network is a collection of _NetworkFeatures. The class inherits the following attributes: Unique ID, Description and name. It is possible for a network to have a sub network, and so it is possible to aggregate different parts of a network with same commodity to a specific network. However the main role of the Network class in the data model is to allow the aggregation of different network features to form a specific network. The IfcSystem entity in IFC, which is a subtype of the entity IfcGroup, represents a similar concept to the Network class in CityGML. Its role is to relate and compose


103

a group of network objects that fulfill a common purpose; it acts as a functional related aggregation of objects, e.g. a system that comprises water distribution elements (pipes or ducts or cables and related items). The IfcSystem did not include any rule to sustain connectivity between the different network objects; however, it is expected that the connectivity that is established between the different network elements will be generated in a systematic way – for example, to establish a flow path (Fig. 4). The attributes of network class, like Unique ID, name and description, can be obtained from the attributes of the IfcSystem entity: Unique ID could be acquired from IfcGloballyUniqueId, name from IfcLable, and Description from IfcText. The complete list of the elements participating in an IFC system can be obtained from the IfcRelAssignsToGroup. Moreover, the IfcRelAssignsToGroup can be used to allow the participation of one element in more than one system. Yet this could be useful for generating the class network link, which allows the creation of the connectivity between different network systems, e.g. a water pump connected to an electricity network. The nesting of a group of network elements to form a sub-network or subsystem can be obtained from the entity IfcRelNests. Figure 5 provides an example of the IfcSystem for a water network; it comprises entities and an alternative representation in UtilityNetworkADE. The parent water system is composed of two subnetworks, hot and cold water; and the network also includes a heating machine which is part of three systems (cold water, hot water, and electricity). The class _NetworkFeature serves as a root class representing any topographical objects of a utility network (e.g. pipes, manholes). The class is derived from the CityGML super class CityObject. Therefore, it inherits all the characteristics of CityObject class. All thematic network classes can be derived from this class, and these thematic classes can be used to provide hierarchy subclasses for division of network objects based on their semantics – that is, their functional meaning in the networks. The network object in class _NetworkFeature is an abstract Feature class in GML. According to the terms of ISO 19109, a feature can have arbitrary attributes, spatial and non-spatial attributes. The spatial attributes serve, for instance, the mapping of actual 3D object geometries in different levels of detail.

IfcSystem

1 ONLY (RelatingSystem) 1 or MANY (RelatedElements) IfcElement

IfcRelAssignsToGroup

Fig. 4 Elements aggregated into a system in IFC (Peachavanish et al. 2006)

104

I. Hijazi et al.

Fig. 5 IfcSystem in IFC vs. Network class in UtilityNetworkADE. Water heater is the Network link which connects the hot and cold water systems

The class _NetworkFeature can be generated using the IfcDistributionFlowElement in IFC (Fig. 6). All building service elements in IFC are exchanged by a subtype of IfcDistributionElement. The various subtypes inherit attributes and relationships from IfcDistributionFlowElement and do not add on any further

Initial Investigations for Modeling Interior Utilities Within 3D Geo Context IFC

IfcFlowTerminal:

105

CityGML

part of the electrical system (terminus), cold water system (terminus) amd hot water system (beginning)

Core::_CityObject

_NetworkFeature

IfcFlowFlowFittings

S-4 S-3

IfcFlowFitting

IfcFlowSegment IfcFlowTerminal

S-2

IfcFlowSegement

S-1

IfcFlowTerminal

Sub-classes derived from the _ NetworkFeature class

Fig. 6 Ifc thematic classes (left) and its alternative representation in UtilityNetworkADE by deriving it from _NetworkFeature class (right)

attributes. These subtypes are introduced solely for the purpose of semantics and logical structuring of the model. IfcDistribuationElement plays a similar role to the _NetworkFeature class. The text below summarizes the subtypes, providing a brief description for their specialized functions. Figure 7 provides an example of these subtypes. l

l

l

l

l

l

l

IfcFlowSegment: defines the occurrence of a segment of a flow distribution system that is typically straight, contiguous and has two ports (e.g. a section of pipe or duct). IfcFlowFitting: The distribution flow element IfcFlowFitting defines a junction or transition in a flow distribution system (e.g. elbow, tee, etc.). Its type is defined by IfcFlowFittingType or its subtypes. IfcFlowTerminals: defines a permanently attached element that acts as a terminus or beginning of a distribution system (e.g. air outlet, drain, water closet, sink, etc.). A terminal is typically the point at which a system interfaces with an external environment. Its type is defined by IfcFlowTerminalType or its subtypes. IfcFlowController: defines a distribution system that is used to regulate flow through a distribution system (e.g. damper, valve, switch, relay). Its type is defined by IfcFlowControllerType or its subtypes. IfcDistributionChamberElement: defines the place at which distribution systems and their constituent elements may be inspected or through which they may travel. IfcFlowStorageDevice: defines a device that participates in a distribution system and is used for temporary storage of a fluid, such as a liquid or a gas (e.g. tank). Its type is defined by IfcFlowStorageDeviceType or its subtypes. IfcFlowTreatmentDevice: defines a device typically used to remove unwanted matter from a fluid, either liquid or gas, and typically participates in a flow distribution system (e.g. air filter). Its type is defined by IfcFlowTreatmentDeviceType or its subtypes.

106

I. Hijazi et al.

IfcFlowStorageDevice

IfcFlowTerminals

IfcFlowFitting

IfcFlowController

Distribution Flow Element

IfcFlowSegment

Fig. 7 Examples for different IfcDistributionElement l

l

IfcFlowMovingDevice: defines an apparatus used to distribute, circulate or perform conveyance of fluids, including liquids and gases, and typically participates in a flow distribution system (e.g. pump, fan). Its type is defined by IfcFlowMovingDeviceType or its subtypes. IfcEnergyConversionDevice is a device used to perform energy conversion or heat transfer and typically participates in a flow distribution system. Its type is defined by IfcEnergyConversionDeviceType or its subtypes.

IFC represent Building service components –IfcDistributionFlowElement using SolidModel representation with its specific types: B-Rep, Surface model and Sweep solid. Transforming geometrical information from the IFC model of these elements in case of sweptsolid requires conversion operations. Special network objects, such as pumps, tanks (IfcFlowMovingDevice IfcFlowStorageDevice), that are composed of several parts can be also represented within _NetworkFeature; this class can consists of one or many network objects. Information about the list of objects that aggregate to form one object can be obtained from the entity IfcRelAggregates. Using _NetworkFeature class, it is possible to aggregate these objects and treat them as though they were a single object.

5.2

Graph Structure

The connectivity between the network objects is given with a graph structure in the UtilityNetworkADE. Each topographic object _NetworkFeature as well as Network class can be associated with graph representation. Similarly, IFC represents


107

the topological relationship between the different network objects, using the concept of connectivity. As mentioned previously, the IFC distinguish two types of connectivity - logical and physical connectivity. In the following text, we will discuss the different classes that comprise the graph in UtilityNetworkADE, and further investigate the related concept in IFC, and the possibilities for transforming it to generate the UtilityNetworkADE graph classes. The class NetworkGraph in UtilityNetworkADE represents the topological view of the utility networks in CityGML. Each thematic network object _NetworkFeature has graph representation using the class FeatureGraph. The graph representation for one feature can be simple - composed of one node derived directly from node class or a composite, where a group of nodes and edges is needed to represent one network feature. In the second case the internal node is distinguished by an interior node (Type: interior) from the node class. Additionally, the connecting links between these nodes are further specified using an interior feature link subtype (InteriorFeatureConnection) derived from the edge class. See Fig. 8 for an illustration of these relations. If two network objects _NetworkFeature (using Feature graph representation) are connected, then their relations can be represented using the NetworkGraph class (Fig. 8). The connectivity is expressed using the class InterFeatureConnection, which is derived from class edge. This holds two attributes: direction, to specify the flow direction in the edge; and link control, which allows controlling the flow in the network so it is possible to control the commodity on the network. The classes that compose the class Network graph are derived directly from the classes Nodes and Edges, which are directly derived from gml::feature and represent a feature in terms of ISO 19109. Its geometry is realized using simple point gml::point and simple curve gml::_curve. In IFC, the concept of connectivity is introduced as a method to represent the relationship between the different IfcDistributionFlowElement. The physical Connectivity concept is generally associated with things that are directly or physically connected. However, there are circumstances where it is important to know that there is a connective relationship between two (or more) objects, but where there is no physical connection to identify this fact. In this case, the idea of non-physical connection is termed logical connectivity, in the sense that there is a logical or implied connection between objects. An example for this connectivity can be seen

Fig. 8 Graph representation in UtilityNetworkADE (composite case), a node type interior is used to connect more than one InteriorFeatureLink (T-fitting) (left) (Becker et al. 2010)

108

I. Hijazi et al.

in a light switch, where the lights are turned on and off by that switch (see Fig. 9). Frequently, the cable that connects the switch with the light will not be instantiated in a model, and so the physical connection cannot be achieved. In this case, the cable making the connection has to be considered. The connectivity information in both cases (physical and logical connectivity with ports) is represented using the same entities. The connectivity information can be extracted from the entity IfcRelConnectsPortToElement and IfcRelConnectsPorts, which has optional attributes, implied the cable making the physical connectivity from switch to light fixture. IfcRelConnectsPortToElement defines the relationship that is made between a port and the IfcDistributionFlowElement. Ports contained in different elements are connected to each other using the IfcRelConnectsPorts relationship. Using both relationships, a topological model can be defined and transformed to direction attributes associated with the edge class in UtilityNetworkADE. Figure 10a–c provides examples representing the connectivity information of network objects in IFC, and the possible transformation to represent them using UtilityNetworkADE classes. The geometry of network graphs in UtilityNetworkADE is represented using node and edge classes, which are derived from GML feature class, which in turn can be associated with simple geometry

Fig. 9 The logical connection between light and light switch key, the route of electrical wire is not available


109

a IFCFlowSegment-Connectivity entities

UtilityNetwork-Featuregraph

InteriorFeatureLink

IfcFlowSegment IfcDistributionPort

Node (type: exterior)

IfcRelConnectsPortToElement #240 = IFCFLOWSEGMENT (‘ABCDEFGHIJKLMNOPQ00040’, #9, ‘C’, ‘ ’, $, $, $, $) ; #242 = IFCDISTRIBUTIONPORT (‘ABCDEFGHIJKLMNOPQ00042’, #9, ‘CO’, ‘ ’ , $, $, $, .SOURCE.) ; #243 = IFCRELCONNECTSPORTTOELEMENT (‘ABCDEFGHIJKLMNOPQ00043’, #9, $, $, #242, #240) ; #244 = IFCDISTRIBUTIONPORT (‘ABCDEFGHIJKLMNOPQ00044’ , #9, ‘C1’, ‘ ’, $, $, $, .SINK.) ; #245 = IFCRELCONNECTSPORTTOELEMENT (‘ABCDEFGHIJKLMNOPQ00045’, #9, $, $, #244, #240) ; #246 = IFCRELDEFINESBYPROPERTIES (‘ABCDEFGHIJKLMNOPQ00046’, #9, $, $, (#240), #247) ;

b IFCFlowFitting-Connectivity entities

UtilityNetwork-Featuregraph

IfcFlowFitting

InteriorFeatureLink

IfcDistributionPort

Node (type: interior)

IfcRelConnectsPortToElement

Node (type: exterior)

#310 = IFCFLOWFITTING (‘ABCDEFGHIJKLMNOPQ00073’, #9, ‘T’, ‘ ’, $, $, $, $); #312 = IFCDISTRIBUTIONPORT (‘ABCDEFGHIJKLMNOPQJ00075’, #9, ‘TO’, ‘ ’ , $, $, $, .SOURCE.); #313 = IFCRELCONNECTSPORTTOELEMENT (‘ABCDEFGHIJKLMNOPQ00076’, #9, $, $, #312, #310); #315 = IFCRELCONNECTSPORTTOELEMENT (‘ABCDEFGHIJKLMNOPQ00078’, #9, $, $, #314, #310); #317 = IFCRELCONNECTSPORTTOELEMENT (‘ABCDEFGHIJKLMNOPQ00080’, #9, $, $, #316, #310);

c Physical connectivity in IFC

UtilityNetwork- NetworkGraph

IfcRelConnectsPorts

InterFeatureLink

#405 = IFCRELCONNECTSPORTS ( ‘ABCDEFGHIJKLMNOPQ00086’ , #9, $, $, #312, #224, $) ;

Fig. 10 (a) IfcFlowSegment associated with two ports modeled using Featuregraph class of UtilityNetworkADE. (b) IfcFlowFitting associated with three ports modeled using FeatureGraph, the interior node is not directly represented in IFC and could be generated on the fly in UtilityNetworkADE. (c) Connectivity relation between IfcFlowFitting and IfcFlowsegment and its corresponding representation in UtilityNetworkADE

representation. IFC allow multiple geometric representations of its elements. Therefore, in cases where the topology representation is available for any IfcDistributionFlowElement entity, it can be transformed into IFC. It should be noted that the representation of network objects as points and lines need harmonization in order to

110

I. Hijazi et al.

Fig. 11 Logical connectivity in case of water tank (Liebich et al. 2010)

have the same classification as the node types (interior and exterior) and edge subtype (InterFeatureLink and InteriorFeatureLink) in UtilityNetworkADE. If the graph representation is not available in IFC (simple line and point of network object), it is still possible to generate it and transform it into the class FeatureGraph with respect to its subtypes in UtilityNetworkADE. Another approach to extract the connection information in case of physical connectivity would be by using the class IfcRelConnectsElements, where each network object has attributes represented by the IfcDistributionFlowElement that this object is connected to or from. Logical connectivity can include cases without ports. It can be seen from the situation of a water supply pipe that discharges into a tank (the particular situation in this example being a potable water pipe, which is taken to discharge into a tank containing non-potable water; in which case, a specific distance between the outlet of the potable discharge and the waterline of the non potable water in the tank must be maintained) (Fig. 11). In this case, the connectivity information is achieved in IFC using IfcRelConnectsElements or IfcRelConnectsWithRealizingElement. Currently, the relation could be modeled in UtilityNetworkADE using the InterFeatureLinks with no geometry representation. However this is not enough and undesirable since it could break the useful concept that unfulfilled port describes an open end in pipe. There is a need for a special class that allows modeling the connectivity and the distance that should be left between the two objects and could represent the air gap.

6 Conclusion The paper investigates the possibilities for integrating interior utility in 3D city models. Following a background literature review, the paper presents an approach for integration interior building utilities in city models by means of semantics


111

mapping. The approach involves harmonizing the semantics and thus allowing formal mapping between the BIM and real world network in UtilityNetworkADE/CityGML (core model). The information provided in the paper can contribute to the efforts of enriching 3D city models with urban knowledge, so as to extend their functionality and usability. The investigation has proved that UtiltiyNetworkADE provides the primary classes, which can be easily extended to model the interior building utilities. Most building service concepts in IFC schema can be mapped to UtilityNetworkADE with lossless of data. Specialized classes can be easily derived from the class _NetworkFeature to represent the different thematic concepts in IfcDistributionFlowElememt. Network Class provide similar concept to the IfcSystem in IFC. The nesting of network objects between different networks can be used to extract the connectivity link (Network Link) between the different network systems. The graph structure in UtilityNetworkADE can be generated using the IFC entities, in case of logical and physical connectivity with ports, connectivity information can be transformed and logical links can be created in UtilityNetworkADE using sub classes of _Edge without having a geometry that represents the physical connectivity. Generating the nodes and edges classes in UtilityNetworkADE with its subtypes in case of physical connectivity requires special consideration; a transformation approach would be needed to tackle problems that can be associated by moving different nodes and edges to the graph classes in UtilityNetworkADE. Some nodes need to be considered as interior nodes and others which represent the ports are considered exterior nodes. It is important to notice that this problem needs harmonizing in all cases of IfcDistributionFlowElememt expect the IfcFlowsegment (e.g. pipes) The IfcFlowsegment is typically represented as a straight element in IFC standard and therefore it represent an Interfeaturelink and the ports represent an exterior nodes. The logical connectivity without ports (e.g. water tank cases) cannot be represented directly in UtilityNetworkADE. An alternative solution could be using Interfeaturelink or Network Link with no geometry representation. However, the solutions don’t address the problem adequately. Although there are explicit semantic classifications on IFC schema regarding interior utilities, it is important here to mention that the investigation shows that the current IFC editors (e.g. AutoCAD MEP, Autodesk Revit MEP, or ArchiCAD MEP) implement the standard in different manners and some concept are implemented but cannot be transferred to IFC standard. For example, the concept of logical connectivity with ports are implemented in AutoCADMEP but cannot be directly transformed to be represented using IFC entities. The ongoing research will focus on formally defining new classes that will help to customize the UtilityNetworkADE to represent the interior Utility, the relation between the UtilityNetworkADE and GeoBIM ADE could be investigated. Moreover, a bidirectional transformation will be considered and investigated, the current work concentrated on unidirectional information transformation (i.e. from a BIM to CityGML models). Such transformation might also be required to support urban upgrading and renovation related tasks where an information model of a building usually does not exit.

112

I. Hijazi et al.

Acknowledgements This paper has been prepared as a part of research project (3D-GIS Campus) in the institute for Geoinformatic and Remote Sensing (IGF), University of Osnabrueck., it is supported by the German Academic Exchange Service (DAAD) under the program: Research Grants for Doctoral Candidates and Young Academics and Scientists.

References Akinci B, Karimi H, Pradhan A, Wu CC, Fichtl G (2008) CAD and GIS interoperability through semantic web services. ITcon 13:39–55. Becker T, Nagel C, Kolbe T (2010) UtilityNetworkADE,-Core Model-Draft version. Retrieved May 3, 2010 from the World Wide Web: http://www.citygmlwiki.org/index.php/CityGML_UtilityNetworkADE Benner J, Geiger A, Leinemann K (2005) Flexible generation of semantic 3D building models. In: Gr€oger G, Kolbe T (eds.), Proceedings of the 1st International Workshop on Next Generation 3D City Models, Bonn, Germany, pp. 17–22. BIM Server (2009) Building information model server/converter to CityGML and KML. Retrieved May 1, 2010 from the World Wide Web:www.BIMserver.org CityGML – ADE (2010), CityGML application domain extensions. Retrieved Mai 1, 2010 from the World Wide Web: http://www.citygmlwiki.org/index.php/CityGML-ADEs Clemen C, Gr€undig L (2006) The industry foundation classes-ready for indoor cadastre? In: Proceedings of XXIII International FIG Congress (eds.), M€unchen, Germany. Emgard KL, Zlatanova S (2008) Design of an integrated 3D information model. In: Coors R, Fendel E, & Zlatanova S (eds.), Urban and regional data management: UDMS annual 2007 (pp. 143–156), Taylor & Francis Group, London, UK. Gr€oger G, Thomas H. Kolbe, Angela Czerwinski, Claus Nagel (2008) OpenGIS® City Geography Markup Language (CityGML) Encoding Standard, Version:1.0.0,OGC08-007r1, http://www. opengeospatial.org/standards/citygml (27 April 2010). Groneman A, Zlatanova S (2009) TOPOSCOPY: a modelling tool for CITYGML. In: Onsrud H, van de Velde R (eds.), GSDI Association, The Netherlands, pp. 1–13. Hijazi I, Ehlers M (2009) 3D web requirement: a case study for the University of Osnabr€uck. In: Proceedings of the 6th International Summit of Digital Earth (CD), Beijing, China. Hijazi I, Ehlers M, Zlatanova S, Isikdag U (2009) IFC to CityGML transformation framework for geo-analysis: a water utility network case. In: de Maeyer P, Neutens T, de Rijck M (Eds.), 3D GeoInfo, Proceedings of the 4th International Workshop on 3D Geo-Information. Ghent University, Ghent, pp. 123–127. IFC Explorer (2008) Tool for viewing and conversion of IFC models. Retrieved May, 20, 2009 from the World Wide Web: http://www.iai.fzk.de/www-extern/index.php?id ¼ 1040&L ¼ 1 Industry Foundation Classes for GIS (IFG) (2009) Retrieved Mai, 20, 2009 from the World Wide Web: http://www.iai.no/ifg/Content/ifg_index.htm Isikdag U, Underwood J, Aouad G (2008). An investigation into the applicability of building information models in geospatial environment in support of site selection and fire response management processes. Advanced Engineering Informatics, 22:504–519. Isikdag U, Zlatanova S (2009) Towards defining a framework for automatic generation of buildings in CityGML using building Information Models. In: Lee J, Zlatanova S (eds.), 3D Geoinformation and Sciences, Springer, Berlin Heidelberg, pp. 79–96. Liebich T, IFC 2x Edition 3, Model Implementation Guide, Version 2.0,http://www.iaitech. org/downloads/accompanyingdocments/guidelines/IFC2x%20Model%20Implementation% 20Guide%20 V2-0b.pdf (13 April 2010). Nagel C, Stadler, A, Kolbe T (2009) Conceptual Requirements for the AutomaticReconstruction of Building Information Models from Uninterpreted 3D Models, Academic Track of Geoweb 2009 Conference, Vancouver.


113

Peachavanish R, Karimi H, Akinci B, Boukamp F (2006) An ontological engineering approach for integrating CAD and GIS in support of infrastructure management. Advanced Engineering Informatics 20(4): 71–88. Safe Software (2008) FME Desktop Translator/Converter Software. Retrieved May 20, 2009 from the World Wide Web: http://www.safe.com/products/desktop/formats.php Schevers H, Mitchell J, Akhurst P, Marchant D, Bull S, McDonald K, Drogemuller R, Linning C (2007) Towards digital facility modelling for Sydney opera house using IFC and semantic web technology, ITcon 12: 347–362. Shen W, et al (2009) Systems integration and collaboration in architecture, engineering, construction, and facilities management: a review. Advanced Engineering Informatics, doi:10.1016/ j.aei.2009.09.001. Wu I, Hsieh S (2007) Transformation from IFC data model to GML data model: methodology and tool development. Journal of the Chinese Institute of Engineers, 30(6): 1085–1090.

.

Depth Perception in Virtual Reality Anja Matatko, J€ urgen Bollmann, and Andreas M€ uller

Abstract Thematic 3D cityscapes represent a combination of the advantages of thematic 2D maps with those of 3D modelling. The application of thematic 3D cityscapes, however, poses the question of difficulties for orientation in the space represented. Here is where, among other things, factors of depth perception must be applied. This study by the University of Trier examines empirically the phenomenon of depth perception in thematic 3D cityscapes to be able to offer recommendations to makers of such models, recommendations for so-called depth cues which improve perception of the spatial environment and which can be technologically implemented. These recommendations are based on theoretical approaches to depth perception which were established in recent years in the field of perception psychology and geovisualisation.

1 Introduction In recent years, development and use of geo-spatial 3D models has increased greatly. Of scientific interest are the processes of acquiring 3D geodata, concepts and methods of landscape or city modelling as well as system development for interactive and stereoscopic model presentation. In addition, research examines problems of acquiring and processing spatially based information and knowledge structures by model users. The great degree of optical immersion found with rendering 3D models leads to the fact that special questions of spatial or depth perception, such as factors for navigation and interactive processes, generally play only a subordinate role. The effectivity of these factors, however, is of interest especially when structures, streets and other model elements are graphically linked to geo-spatial problems, which is the case in Virtual City Models. The term “Virtual City Model” covers thematic 2D maps, block models, photorealistic models and thematic 3D

A. Matatko (*), J. Bollmann, and A. M€ uller Cartography, University of Trier, Behringstrasse 54286, Trier, Germany e-mail: [email protected], [email protected], [email protected]


115

116

A. Matatko et al.

city models. The neglect of rendered elements in thematic 3D city models, such as object textures and colours leads to a reduction of possibilities for mental identification, orientation and recall. This situation can be seen in a thematic city model developed at the University of Trier, one in which rendered elements were greatly reduced: the model is described in Matatko et al. (2009, 2009). Figure 1 shows a detail of the model. Streets, houses and roofs represent classified thematic attributes. Such models can especially be used for urban development planning. In order to make mental processing of information in such models as effective as possible, however, our aim is to analyse the entire field of spatial and depth perception and study factors which can be integrated, if necessary, into thematic city models. Some elements leading to depth perception are already implemented in the construction and programming of 3D models. The person using the 3D modelling and rendering software must also, if necessary, conceptualise and integrate additional elements. Thus, the question arises which elements facilitating the impression of spatial depth actually possess relevance in cartographically oriented 3D models. Our aim is to get information about the significance of each single depth cue. As most of the software automatically integrates several depth cues, we tried to isolate them as good as possible in order to get empirical evidences for their significance. In this article, we initially discuss different concepts of depth perception and its components. Then, we focus on selected results of a comprehensive empirical study. A larger goal is to clarify relationships and degrees of effectivity as well as the possible applications of factors of depth perception when using virtual cityscapes.

Fig. 1 Thematic 3D city model created by the University of Trier with representation of the living quality index as well as the public space quality index

Depth Perception in Virtual Reality

117

2 Depth Perception Examining spatial perception and its properties occupies an essential part of research into the psychology of human perception. Figure 2 represents the study topic in the psychology of perception according to Guski (1996), stating that all sense organs are relevant to acquiring information relevant to space. Not only visual perception but also the other sense organs lead to an impression of spatial depth, especially hearing. This study only treats questions of visual perception. Goldstein (2002) differentiated between two basic theoretical approaches to explain depth perception: the cue theory and the ecological approach proposed by J. J. Gibson (1973), with the cue theory following the constructivistic approach. It assumes that the observer plays an active role in the perception process. The cue theory intends to identify information on the retinal image, processed information matching the depth of the actual, existing world. The ecological approach emphasises the observer and his/her interaction with the environment. This chapter is based on different schemata of depth cues, as proposed by Goldstein (2002), Albertz (1997), and Ware (2004). As there is no generally accepted schema about depth cues and their impact on depth perception (and our aim is not to create one), we summarize different depth cues and depth perception gradients from different authors.

2.1

The Basic Approach of the Theory of Depth Cues

The theory of depth cues and related unconscious information processing was proposed by Helmholtz as early as 1896 (von Helmholtz 1896). According to Albertz (1997), the third dimension within the visual perception process is “constructed”

Fig. 2 Study topics in the psychology of perception (Altered on the basis of Guski 1996)

118

A. Matatko et al.

from the signals arriving on the retina. This process requires several elements of spatial perception to achieve depth perception. The impression of depth arises through combining experience, environmental stimuli and images of these stimuli on the retina. These elements, called depth cues, may, as proposed by Goldstein (2002), be divided into four groups: oculomotor cues, pictorial cues, motion-produced cues and binocular disparity. Other groupings differentiate between binocular and monocular cues. In Goldstein, the first three groups comprise monocular cues, that is, they can also be processed with only one eye. 2.1.1

Oculomotor Cues

In contrast to other depth cues, oculomotor cues are perceptible because of the alterations in the eye during movement of the eye as it perceives space. In fixating on a point in space, the lens of the human eye becomes curved and accommodates. The further the eye moves its focus from the fixation point, the more blurred the image becomes. Spatial depth can thus be estimated. However, the observer cannot recognise whether objects lie in front of or behind the fixation point. With objects far removed from the fixation point, the effect of the blur becomes weaker for optical reasons. Goldstein (2002) numbers convergence and accommodation amongthe oculomotor cues. He defines convergence as the movement of muscles causing an inward movement of the eyes. He defines accommodation as the bulging of the lens while focusing on near objects. Kelle (1995) considers accommodation subordinate, as he views the reaction of the eye muscles as one based on an already realised depth perception. 2.1.2

Pictorial Cues

Pictorial cues comprise the largest share of monocular cues. They are based on interpretations of information derived from images. As the retina itself receives only a two-dimensional copy of reality, the observer must use optical links to obtain a three-dimensional impression. Six pictorial cues are introduced in the following: l

l

l

l

l

Overlap. A topological alignment of depth can be created when objects further in the background are occluded by those closer to the observer. Size in the field of view. The greater the portion of the object’s surface is in the field of vision, the larger the object will be estimated to be. Height in the field of view. Objects located higher up in the field of vision are perceived to be further away. Atmospheric perspective. Contrast reduction with very great distances, owing to particles in the air (“fogging effect”). Familiar size. The observer knows from experience that objects become smaller the further away they are. If either the object size or the distance is known, the observer can draw corresponding conclusions concerning the unknown parameter.

Depth Perception in Virtual Reality l

119

Linear perspective. Parallel lines in space converge toward a common vanishing point, closely related to texture gradient.

2.1.3

Motion-Produced Cues

Besides depth cues available to a static observer, other depth cues exist which become effective only when the observer moves. l

l

Motion parallax. Near objects appear blurred to the observer and move quickly, whereas objects further away move more slowly and are more easily identified. Albertz (1997) accorded motion parallax “enormous importance for perceiving our environment”. Deletion and accretion. In non-perpendicular movements, objects not at the same distance tend to appear to move relative to one another. This effect is called deletion when the object in the background becomes increasingly occluded by the object in front of it. It is called accretion when the object in the rear emerges from behind the object in front.

2.1.4

Binocular Disparity

Binocular disparity is the most important cue for depth. It is defined as the acquisition of a spatial impression through the effect of binocular disparity, an effect arising from the off-set position of the two human eyes. Due to the different positions, images are created on both retinas from two different perspectives. In 1959, Be´la Julesz was able to confirm binocular disparity with the help of his Ramdom Dot Stereogrammes, in which a spatial impression could be created without using further depth cues.

2.1.5

Effectivity of Depth Cues

The individual depth cues develop their effectivity at different distances. For example, atmospheric perspective is effective only at distances greater than 30 m (100 ft), whereas the effect of binocular disparity is already greatly reduced after a few metres, and, according to Buchroithner and Schenkel (2001) ceases completely at 10 m. Further effects are represented in Table 1.

2.2

Depth Perception Gradients

According to Gibson (1973), the environment consists of stimuli which can be received and processed by living things according to their abilities. The author lists several requirements for research into depth perception: First, depth perception

120

A. Matatko et al.

Table 1 Effectivity of selected depth cues (Altered on the basis of Goldstein 2002)

Depth cue Overlap Familiar size Oculomotor cues Binocular disparity Atmospheric perspective

30 m No Yes No No Yes

should be examined by a moving observer, as dynamics prevail in reality. Second, in contrast to the theory of depth cues, it is not the image on the retina to be analysed but the information available to the optic array. Gibson (1973) defines various elements of optic arrays similar to depth cues, elements leading to depth perception, among them texture gradients. Albertz (1997) expanded and differentiated the effect and the significance of texture gradients, which are an optical transitional phenomenon continually extending into spatial depth. The author thus differentiates size, density and form gradients, all of which are interrelated to the conditions of the central perspective structure of space. For example, windows leading off to the vanishing point appear to be continually reduced in their size and intervals as well as to systematically be altered in form, thus creating a gradual but continuous gradation. Closely related to this phenomenon is the actual texture gradient, in which, with increasing spatial depth, the grain of the surface pattern, for example, a cobbled street or vegetation, becomes optically finer. Additional gradients are the contrast and colour gradients arising from atmospheric phenomena as characteristics of the so-called atmospheric perspective. According to Albertz (1997), continuous brightness gradients created by the formation of cast shadows on objects have great significance. Sequences of projected views of objects extending into spatial depth can result through increasing gradual distortion and deformation through cast shadows and reflections.

2.3

Status of Current Research in Cartography and Geovisualisation

Thus far, hardly any concrete and complete empirical studies on depth perception have been published in the fields of cartography or geovisualisation. Especially informative for this study, however, are the publications of Albertz (1997) and Buchroithner (2001). The latter work differentiates various process groups creating spatial images which differ in the depth cues used, resulting in degrees of perception and immersion. The authors stipulate that for variable visualisation tasks of geodata the individual processes must be evaluated on the basis of test applications. They introduce a selection of processes based on estimates and point to continuing studies still required for evaluation of this pre-selection.


121

As mentioned above, Albertz (1997) expressly differentiates among the effects of various perception gradients and places them in the context of the active observer in the perception space. Thus taken into consideration, on the one hand, are the properties of optical processes into spatial depth known from the observer’s environment; and addressed, on the other hand, are the active positioning as well as mental attitude and direction of the observer. Both aspects are of special interest for the user-oriented, active mastering of virtual space. Kraak (1988), Meng (2002), Jobst and Germanchis (2007), for example, all require an ample number of depth cues as well as sufficient a priori knowledge for the observer to be able to acquire a spatial impression. On the whole, an acceptable theoretical basis resulted in being able to develop the tasks in the following study and to classify the empirical conclusions.

3 Empirical Analysis of Selected Depth Cues 3.1

Methodical Background

The empirical study was conducted with the aid of the 3D model of the Trier inner city created by the cartography department. In ten studies, eight depth cues were examined, represented in Table 2. Along with the above-named theoretically documented factors, additional influential factors were examined experimentally, factors from the field of spatial representation and perception. A prerequisite to studying a depth cue is that it both had to be implementable with the existing technological capabilities and had to be regarded as relevant also for thematic 3D models. The software used for modelling and navigation in the 3D models was a combination of 3D Studio Max and Virtools. The latter facilitates creating applicationbased programming of navigation and interaction functions based on a modular system. The main problem in the studies was that the individual depth cues were difficult to isolate from one another, as several depth cues normally exist automatically. This was especially true of the study of overlap. Each study had a sample size between 25 (for those using eye tracking) and 200 (for those using online surveys).

Table 2 Studies of depth perception in 3D city models

Depth cues Binocular disparity Linear perspective Motion parallax Overlap Texture gradient Contrast gradient Depth of focus Viewing angle

Implementation Time consuming Automatic Automatic with dynamics (movement) Automatic Automatic Integrable Complex Automatic

122

A. Matatko et al.

Table 3 Participant characteristics of selected studies of depth perception Study Texture gradient Depth of field Contrast gradient Number of Online survey: Online survey: 400 Eye tracking: participants 200 students 40 students Eye tracking: 20 students Intellectual n.a. Non-experienced n.a. background and experienced pre-experience computer users

Motion parallax Eye tracking: 28 students

Non-experienced and experienced VR-users

Table 2 represents the software implementation capabilities of all the examined depth cues. In the following sections, four studies will be presented in detail: texture gradient, depth of field, contrast gradient and motion parallax. Table 3 compares the exact number of study participants, their intellectual background and preexperience.

3.2

Texture Gradient

In their study, Bott et al. (2007) have documented the effect of texture gradients empirically. Each task contains distance estimations. The distances are to be given as relative or absolute but with an indicated reference distance. The ground surface for the study is represented in three different scenes: without texture, without texture gradient (distorted texture with consistent grain; that is, the intervals toward the rear do not become smaller), with texture gradient. To avoid perspective influences, neither an unstructured asphalt surface nor a paved structure with regularly ordered stone was used. The results of the study are displayed in Fig. 3. The largest percentage of correct answers (73%) occurred with scenes showing undistorted texture (that is, with natural texture gradient). The response ratio was about identical for the scenes using the distorted texture or no texture, with approximately 30% accurate estimations. With the aid of eye tracking, it could be shown that the test persons oriented themselves on the ground texture in the landscape to estimate the distances. “Hot spot” images were used to illustrate the eye movements. Figure 4 contrasts the results of a scene with and without ground texture. In the scene without texture, it can clearly be seen that the test persons sought other orientation points in the landscape to be able to answer the question. In the scene with ground texture, the fixations are concentrated on the central area of the surrounding ground texture. The behaviour of test person shown in Fig. 4 is representative for all test persons participating in the study of texture gradient.


123 Results from all scenes

Accurate estimations (percent)

60

54,2

50 40 30,6 27,0

30 20 10 0 without texture

normal texture

distorted texture

Fig. 3 Results from the test images with texture, without texture and with distorted texture (Altered on the basis of Bott et al. 2007)

Fig. 4 Hot spot analysis of scenes without (left) and with (right) texture (Altered on the basis of Bott et al. 2007)

3.3

Depth of Field

Depth of field is a phenomenon that plays a role primarily in photography and video recording techniques, in which an emphasis is placed on a desired section of an image by setting the lens for sharp or unsharp areas. Greiling and Marx (2007) studied the effect of depth of field created through optical structuring of image scenes in city models as a possible factor in depth perception. Test persons had to solve tasks within several scenes with different depth of field settings. Figure 5 shows two details of the study. As part of an Internet survey, settings and acceptance criteria were ascertained in addition to the experiment. The results of the study were surprising in that, with depth of field, subjective impression and objective action were widely divergent. In the survey, 90% of the test persons declared that they preferred scenes with sharp depth of field.

124

A. Matatko et al.

Fig. 5 Landscape completely sharp (left) and with foreground focus (right) (Altered on the basis of Greiling and Marx 2007) Correct answers according to scene variation


90 80 70 60 50 40 30 20 10 0 small dof, large dof, small dof, large dof, small dof, large dof, small dof, large dof, large area, large area, large area, large area, small area, small area, small area, small area, fore focus fore focus back focus back focus fore focus fore focus back focus back focus

no dof

Fig. 6 Success ratio with variations in depth of field, size of area and kind of focus (dof ¼ depth of field) (Altered on the basis of Greiling and Marx 2007)

Figure 6 shows, however, that the success rate for scenes with focus in the foreground rises and thus probably supports the solution to the tasks through effects of depth of field. There are few differences between men and women. There is also no significant difference in the comparison between larger and smaller areas.

3.4

Contrast Gradient

Voshaar and Metzger (2007) have represented the effect of contrast gradients. Combined with colour gradient, it forms the atmospheric perspective. For technical reasons, only the contrast gradient was studied, as the colour gradient is not implemented in Virtools. During the study, the fixation movements of the test persons were recorded and evaluated using the Tobii Eye Tracker. Details from


125

the city model were shown at different positions. The tasks related to estimating distances of objects to one another, that is, test persons were asked to estimate relative values instead of absolute distances. Figure 7 represents two scenes from the study: without decreased contrast on the left and with decreased contrast on the right due to haze. The evaluation proved to be a problem because, in general, the estimations were grossly incorrect, regardless of the amount of contrast decrease. Figure 8 shows that, with a great amount of haze, the percentage of correct estimations clearly increased (45%) in comparison to the scenes without or with little haze. Information about spatial orientation can also be obtained from the parameters of eye movement. In the scenes without decreased contrast due to haze, more fixations were counted, ones which were shorter, on the average, than in the scenes with decreased contrast. With the example of one scene, Fig. 9 represents the average fixation time. Based on the shorter fixation time, we can conclude that the test persons were better able to orient themselves in the scene with decreased

Fig. 7 Landscape without (left) and with (right) decreasing contrast due to haze (Altered on the basis of Voshaar and Metzger 2007)


Results according to amount of haze 50.0 45.0 40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0

45.0

29.7

no haze

26.9

medium haze

dense haze

Fig. 8 Evaluation of the results of estimations according to amount of haze (Altered on the basis of Voshaar and Metzger 2007)

126

A. Matatko et al.

Average fixation duration (ms)

Average fixation count (mean) 225

222

220 214

215 210 205

204

200 195 no haze

medium haze

dense haze

Fig. 9 Average fixation time with scenes with greatly different contrast effect (Altered on the basis of Voshaar and Metzger 2007)

contrast. The eye movements occur less aimlessly. Objects were fixated longer, indicating that more exact information gathering was possible.

3.5

Motion Parallax

To find out whether motion parallax increased depth perception, the test persons were to estimate the distances of objects. Glass recycling bins were selected as actual objects from the cityscape; the effects of other cues such as cast shadows or overlaps were limited by configuring the parameters in Virtools. Scenes with and without movement were juxtaposed. Each movement occurred parallel to the objects, at two different speeds. The results determined that no significant difference exists between the slow and the fast camera movement. However, in respect to the number of correct estimations, distinctly appreciable differences exist between the static and the dynamic image. The mean percentage of correctly solved tasks lay higher with slower or faster camera movement than for the static view. In addition, the scatter was less, especially with the slow camera movement, than with the use of screenshots. It can thus be concluded that motion parallax contributes essentially to depth perception in the thematic landscape.

3.6 3.6.1

Further Study Results Binocular Disparity

In the study of binocular disparity, difficulties occurred because this disparity, as described above, is effective only at close range. Thus the question arises whether


127

stereoscopy is of great significance at all for perception in a thematic 3D cityscape. This applies primarily to mental processes in which acquiring a spatial overview and the mental construction of spatial coherence is emphasised.

3.6.2

Linear Perspective

If central perspective is exchanged for parallel perspective, the exchange has a negative effect on achieving perception. Sex-related differences exist insofar as the answers of men were more correct than those of women. Similarly, positive correlations exist between the prior experience with 3D models and the correctness of answers. With a dynamic representation, the test persons were able to alter the viewing position in such a way that they could select between parallel perspective and central perspective. The test persons preferred the central perspective.

3.6.3

Overlap

In the study of overlap, both the number of objects to be identified and the number of overlapped objects was varied. Fixations were recorded using eye tracking. It was shown that, as the number of objects in a landscape to be identified increased, the response time rose. No significant results were achieved in respect to variation in the number of the overlapped objects, a result associated with the fact that, among other things, the effect of overlap can be substantiated only with difficulty, as this depth cue can hardly be observed isolated from others. The visual use of overlapping objects could, however, be documented with the aid of hot-spot analyses of the fixations.

3.6.4

Viewing Angle

In 3D models, the viewing angle can be varied at liberty. Thus the question arises which viewing angle can be considered ideal. This applies especially to models where no completely free navigation exists but where, for example, aerial tours are conducted along a defined path. The study was able to document that an angle of approximately 15 was considered ideal.

4 Conclusion This study documents the relevance of individual depth cues in spatial perception in thematic 3D models. The isolation of individual depth cues, however, proved to be a problem for the study, for example, with overlap and linear perspective, as these occur of necessity in 3D models where it is difficult to limit them graphically as

128

A. Matatko et al.

well as in their effect. In general, it was attempted to isolate the studied depth cues as far as possible. A further step in the study was to conduct a classification of depth cues according to their relative importance to assist the makers of 3D cityscapes in creating better user-oriented models, so that such models can simplify orientation and mental information processing by the user. It was shown that contrast gradient, motion parallax and texture gradient are especially effective for the concrete application of the Trier 3D city model. However, the results must be further examined, in particular with the increase of free navigation in the 3D models, and adapted to new technological possibilities and their ensuing conditions of perception in models.

References Albertz, J.: Die dritte Dimension – Elemente der r€aumlichen Wahrnehmung. In: Albertz. J. (ed.): Wahrnehmung und Wirklichkeit. Schriftenreihe der Freien Akademie, vol. 17, pp. 81–108. Freie Akademie, Berlin (1997) Bott, F., Mertes, N., Lutz, R.: Empirische Untersuchung zur Wirkung des Texturgradienten anhand einer 3D-modellierten Szene aus dem Stadtteil Trier-Nord. Unpublished project report, University of Trier, Trier (2007) Buchroithner, M.F., Schenkel, R.: 3D-Visualisierung von Geodaten – perzeptionstheoretische und pr€asentationstechnische Grundlagen. In: Koch, W.G. (ed.): Theorie 2000. Vortr€age des kartographischen Symposiums am 17/18 November 2000 an der TU Dresden. Kartographische Bausteine, vol. 19, pp. 113–120. Institut f€ ur Kartographie, TU Dresden, Dresden (2001) Gibson, J.J.: Die Sinne und der Prozeß der Wahrnehmung. Verlag Hans Huber, Bern, Stuttgart, Vienna (1973) [The English original is titled: The Senses Considered as Perceptual Systems. Houghton Mifflin, Boston (1966)] Goldstein, E.B.: Sensation and Perception. 6th ed., Wadsworth Group, Pacific Grove (2002) Greiling, M., Marx, A.: Projektbericht zur Untersuchung des Tiefenfaktors ‘Tiefen(un)sch€arfe’ auf seine Relevanz bei der Tiefenwahrnehmung des Menschen anhand eines Online-Experiments. Unpublished project report, University of Trier, Trier (2007) Guski, R.: Wahrnehmen. Eine Einf€ uhrung in die Psychologie der menschlichen Informationsaufnahme. Kohlhammer, Stuttgart (1996) Jobst, M., Germanchis, T.: The Employment of 3D in Cartography – An Overview. In: Cartwright, W., Peterson, M.P., Gartner, G. (eds.): Multimedia Cartography, 2nd ed., pp. 217–228. Springer, Berlin Heidelberg (2007) Kelle, O.: Dynamische Tiefenwahrnehmungskriterien in Computergenerierten Interaktiven Szene und Virtuellen Simulationsumgebungen. VDI-Verlag GmbH, D€usseldorf (1995) Kraak, M.-J.: Computer-Assisted Cartographical Three-Dimensional Imaging Techniques. Delft University Press, Delft (1988) Matatko, A., M€uller, A., Bollmann, J.: Ein Wohnumfeldqualit€atsindex f€ur Trier. In: Strobl, J., Blaschke, T., Griesebner, G. (eds.): Angewandte Geoinformatik 2009, pp. 826–835. Wichmann, Heidelberg (2009) Matatko, A., M€uller, A., Bollmann, J.: Thematische 3D-Stadtmodelle: Modellierung, Anwendung und empirische Befunde. In: Mitteilungen aus dem BKG, Band 45: Arbeitsgruppe Automation in der Kartographie – Tagung (2009), S99–S106 Meng, L.: How can 3D Geovisualization Please Users Eyes Better? Geoinformatics Magazine for Geo-IT Professionals, vol. 5, 34–35 (2002)


129

von Helmholtz, H.: Handbuch der Physiologischen Optik. 2nd ed., Verlag von Leopold Voss, Hamburg (1896) Voshaar, E., Metzger, M.: Bericht zur kartographischen Projektstudie. Unpublished project report, University of Trier, Trier (2007) Ware, C.: Information Visualization. Perception for Design. 2nd ed., Morgan Kaufmann, San Francisco (2004)

.

Interactive Urban and Forest Fire Simulation with Extinguishment Support ´ lvaro Segura, Anis Korchi, Jorge Posada, and Oihana Otaegui Aitor Moreno, A

Abstract Fires and other related disasters provoke great destruction of high valuable environments and economical losses, especially when they are located in urban areas. In this work, we present a combined urban and forest fire spreading algorithm to be used in real time and interactive virtual simulations. The algorithm is pedagogical oriented and its purpose is not focused in achieving precise results that could be used to predict the fire evolution. The main objective is to obtain a fast, interactive and quasi-realistic virtual simulation to be used in the simulation of virtual scenarios where firefighters and controllers will be trained. The algorithm supports the main variables involved in the fire spreading (slope and wind) and the radiation effect. An additional method has been added to extinguish the fire.

1 Introduction Fire is one of the most complex and destructive phenomena in Nature. When they are out of control, they can devastate large extensions of forest area or burn buildings provoking economical losses, environmental impacts and even human casualties (see Fig. 1). The preventive measures are very important, but eventually, the fire will start. Whether in urban or in forest areas, we can stop or limit a fire by having a skilled and experienced Fire-Force, from the firemen to the management staff who will organize the available resources to fight the fire. Emergency Training Centres are a key factor to successful fighting against fire. The firemen and controllers learn all the relevant aspects, theoretical concepts and methodologies. The practical knowledge is achieved thanks to controlled real firefighting exercises or drills (see Fig. 2). But, due to obvious limitations, the training centres can’t recreate all the possible scenarios (a forest fire with specific

´ Segura, A. Korchi, J. Posada, and O. Otaegui A. Moreno (*), A Vicomtech, Mikeletegi Pasealekua 57, 20009 San Sebastian, Spain e-mail: [email protected], [email protected], [email protected], jposada@ vicomtech.org, [email protected]


131

132

A. Moreno et al.

Burnt Areas in ha in Southern Europe (1980 - 2006)

Italy 24 %

Spain 38 %

Portugal 23 % Greece 9% France 6%

Spain

5.070.305

Italy

3.128.592

Portugal Greece

3.121.776 1.167.396

France

810.417

Total

13.298.486

Fig. 1 Burnt areas in hectares in Southern Europe between 1980 and 2006 (Forest Fires 2010; Schmuck 2006)

Fig. 2 Aircraft rescue and firefighting basic training (Rocky Mountain website 2010)

weather conditions, a city scale urban environment, etc.) because they are too risky and firefighters could be injured during the training. The most suitable solution to fill this gap is the use of virtual simulations, where the firefighters and controllers can safely collaborate to fight a virtual fire and check if the concepts learnt in the training centre have been applied correctly. In order to create high quality virtual simulations, two main aspects must be fulfilled: (1) a good reconstruction of the control centre and a proper emulation of the communication process that will be operated by the controllers to make the decisions about the correct deployment of available resources, and (2) a highly interactive and collaborative virtual simulation for firefighters, controllers, where all the available resources must be represented and operated to extinguish or limit a virtual fire and also, protect the civil people.

Interactive Urban and Forest Fire Simulation with Extinguishment Support

133

One of the main algorithmic elements in such virtual simulations is to simulate how fire spreads as simulation time advances. If a very unrealistic or simplistic algorithm is used, the simulation could not be precise enough. This makes the virtual fire behaviour totally predictable during training sessions, which reduce the pedagogical effect. In contrast, introducing the most complex mathematical model in the simulation will increase the realism of the spreading algorithm, but a real time implementation will be less feasible and will also reduce the user interactivity required for a virtual simulation. Although a simplification of the algorithms is needed in order to achieve interactive rates, it is desirable that it could deal with different types of forest and urban fires react to different types of terrain and buildings, slopes and changing weather conditions. The simulation should support the capability of fire to be extinguished by itself (fuel combustion) or by the firefighters (fire extinguish agent). In this work, we present a fire spreading algorithm that can be implemented to be used in real time and interactive virtual simulations. The algorithm is pedagogically oriented and its purpose is not to produce accurate results that could be used in scientific or engineering applications. The main objective is to obtain a fast, interactive and quasi-realistic system to be used in the simulation of virtual scenarios where firefighters and controllers will be trained. In the next section, the main fire spreading algorithms will be reviewed. Next, the proposed algorithm and methodology will be introduced, followed by some results. Finally, the conclusions and future work will be addressed.

2 State of the Art There are two major models of fire simulation, i.e., empirical models and physical models. The empirical models are made thanks to the experience of real fire, i.e., those models use statistical relationships found between the fire evolution and different parameter tested on the field (Rothermel 1972). In this case, we can mention FARSITE (Finney 1998), which use Huygens principle of wave propagation. The second type, physical-based models, use convection and heat transfer mechanism, but also computational fluid dynamics methods. The main mathematical tools used here are partial differential equations and reaction diffusion systems. Fire Dynamic Simulator (FDS, National Institute of Standards and Technology – NIST) or FIRETEC (Linn et al. 2002) follow this approach. The advantage of those models is their accuracy in the fire prediction. But the computational effort is very high. The mathematical models are too complex and computers can only provide approximate solutions (Dumond 2008). Another consequence of the model’s complexity is that the spatial resolution required is too high. Unlike the two previous models, other research works have taken a different direction from the complex mathematical models. Their objective is to reduce computation time and to implement a real time simulation.

134

A. Moreno et al.

Gary L. Achtemeier (2003) presented the Rabbit Model, a collection of basic rules of fire evolution, which are implemented as autonomous agents (the rabbits). The scope of the Rabbit Model is limited to the evolution of forest fire. Hamada (1951) is one of the first researchers in urban fire models. His model provided empirical equation describing the speed of the fire spread depending on the wind speed and direction. Hamada’s model defines a field compound by identical building blocks separated with the same distance and the fire spread has an elliptical shape (Scawthorn et al. 2005). More recently, some physics based models were made (Lee et al. 2008). This type of model gives more accuracy in the simulation, using equations to describe the heat transfer between buildings (radiation, convection), the temperature modification, and flame shape (direction, length) coming out of building (Iwami et al. 2004) through the windows. The physics based models are also used to simulate the fire spread in nonhomogeneous cities (on contrary with Hamada’s model) with a higher resolution (Weise and Biging 1996). Other physical based models were proposed using cellular automata with a 9 m2 grid cells (Ohgai et al. 2005) or a vector based approach in which each building is a vector object (Iwami et al. 2004; Tanaka and Himoto 2006). One limitation of both models is that they limit the fire spread to entire buildings or individual floors, not taking into account the inner structure of the buildings (rooms). The proposed algorithm in this work presents an urban and forest fire spreading simulation, whose main characteristics are: l

l

l

The fire evolution in forest areas is based on the terrain topology, material and weather conditions. In urban areas, the different buildings characteristics are taking into account to simulate more accurate results. The fire is allowed to cross rivers, firewall or other barriers by introducing the radiation effect. The radiation effect is also introduced to allow the fire spread between near buildings. In a similar way, urban fire can spread to forest areas and vice versa. Very low complexity, allowing real time simulation even with standard computing power.

In the next sections, the details of the proposed method are explained, showing some basic results of the implementation of the algorithm. First, the general approximation of the forest fire and the field definition is introduced. Next, this model is extended with the urban fire model, followed by the extinguishment support.

3 Proposed Method for Forest Fire Spread Simulation The proposed algorithm main goal is to simulate how the fire spreads in urban and forest environment under different circumstances. In order to explain the method, the field where the fire will be must be defined.


3.1

135

Introduction and Field Definition

The forest fire simulation method is based on a divided field in autonomous cells. The field is regularly divided in a grid of square cells and is defined by its geometrical information (origin, size and elevation). Every square cell contains all the necessary information to support the fire spreading, being divided in two main categories: (1) static information, which characterizes the cells and will be used in the algorithm rules, and (2) runtime information, which comprises all the information that will be modified or calculated during the simulation. The static information of each cell corresponds with the geometrical data (position, size and altitude of the cell in the field) and the characterization of the cell (semantics), i.e., type of cell which will determine the nature of the cell (dry grass, tall trees. . .) and the starting quantity of combustible The fire spreading algorithm is a step-by-step process, i.e., every time that the algorithm is run, a new state of the fire evolution is calculated taking into account the elapsed time. The runtime information will be continuously calculated during each simulation step, and includes some important variables like the quantity of combustible, the power or intensity of the fire and the state of the fire. All this information is required in order to produce the final rendering of the fire in the virtual simulations. Additional information to support the fire extinguish method are required, but they will be introduced in the corresponding section.

3.2

Basics Elements of the Fire Spreading Algorithm

The objective is to reduce the algorithmic complexity and the processing time while keeping a correct behaviour of the spreading of the fire. To reach this goal, among the whole set of variables and physical parameters which can influence the fire behaviour, we have chosen the most significant ones. The most relevant external parameters are the wind and the slope (Weise and Biging 1996). Other variables, although necessary for precise simulations or predictions have been discarded. All the cells in the field have an internal state which corresponds with the state of the fire that exists in such cell. The different states are Safe, Activated, Burnt, Survived, RiverCrossing and FireStopped (see Fig. 3). When a cell is in the Safe state, there is no fire in it. The Activated state indicates that there is an active fire in the cell. The Burnt state is the final state of a cell after being completely consumed by the fire. The Survived state is pseudo-final state, when the cell is burnt, but still has residual heat. Even being completely consumed by the fire, the cell can irradiate some heat to others cells. Eventually, the cell will pass to the Burnt state. The RiverCrossing state is a hidden state used in the computation of the fire spread through a river or firewall, using the radiation effect.

136

A. Moreno et al.

Fire Stopped

Activated Safe (t == 0)

Burnt RiverCrossing

survive

Fig. 3 The different states of a cell and the available transitions

Finally, the FireStopped state represents an intermediate state of a cell, when some external fire extinguisher has stopped the fire. The algorithm that defines how the fire spreads in wild land is based on the cells’ initial data of the whole field (combustible, states . . .). Following a similar approach to Gary L. Achtemeier in the Rabbit Model (Achtemeier 2003), we define the fire spread as a displacement of mice. The combustible on the field is considered as the mice’s food (cheese). Thus, the basic concept is that when a mouse eats a cheese, it is equivalent to when a combustible unit is burnt by the fire. A fire is started by positioning a mouse in a square cell, changing its corresponding internal state to Activated. The added mouse will interact with the neighbour cells following some rules. Every mouse is born in a square with a given power to eat a quantity of cheese per simulation step (fire intensity), which changes depending on the type of the cell. When a mouse eats a given quantity of cheese, its power is increased in the same amount, which simulates how the fire intensity is continuously growing while there is available combustible. If there is no cheese remaining, the mouse dies. When a mouse dies, up to eight new mice can be born, since there are eight potential neighbour squares. The algorithm uses the local slope and wind in order to determine which neighbours will be the destiny of the new born mice. The mouse will give birth to all the squares which are in an area of 45 of the wind angle. If there is no wind (or it is too slow), the mouse will give birth to other mice in all the eight neighbours. The mouse will always try to give birth to other mice in squares that are in higher altitude. This behaviour simulates the fact that the fire goes up if there is a slope. Only if there is not a non-burned square at a higher altitude, the mouse can go down, as it has no option to go up. A square can be the destination of multiple mice births, as long as it contains cheese. Thus, a square can have more than one mouse, but the food contained in the square will be more quickly eaten, since their eating power is summed up.


137

Every square cell has a parameter named Fire Power, which is in fact the sum of the mice’s power of the cell. Similarly, the Fire Power of a given cell will be stronger or more intense with a higher number of mice.

3.3

Elliptical Shape Support

The Huygens implementation (Richards and Bryce 1995) uses differential equations to describe the expansion of an elliptical wave. As the fire evolution is calculated, a typical elliptic shape is obtained. In this method, the ellipse is simulated by combining the effect of the slope and wind and giving a probability to the mice to give birth in a wrong direction (see Fig. 4). Those new mice (named Black Mice) reduce their power in each step, instead of the normal behaviour. The consequence is that the fire will evolve very slowly in the wrong direction.

3.4

Radiation Effect

The radiation effect is added to support the fact that a fire can cross a river, firewall or other barriers like roads. It is based on the heat radiation. In the proposed algorithm, a fire is stopped when it encounters a barrier. Depending on the Fire Power in the area, and considering the width and type (streets, road, river . . .) of the barrier, it can be overridden. The fire is able to bypass the river only if the existing vegetation is composed of trees or any other high vegetation. The main idea is the difference of size: the more a tree is high, the more the wind can spread heat/radiation to the other side of the barrier.

Fig. 4 Elliptical shape of a fire in a flat scenario. The wind goes from top-left corner to bottomright corner. Black cells are totally burnt, red ones are activated

138

A. Moreno et al.

Fig. 5 The fire has bypassed successfully a river. The sticks in the water represent the Fire Power and it can be seen that it is being reduced gradually. In the presented case, the fire reached the opposite bank, spreading it

As the river is part of the field, it is also divided in cells (Water Squares). The spread of the fire in those squares follows the wind rule, but mice in those squares don’t eat combustible so the Fire Power is never increased. Moreover, when a mouse born in water squares, the Fire Power is not raised in that square but decreased, i.e., the new mouse will have less power than its mother (see Fig. 5). As the fire is stopped when the power is zero or negative, there are not too many possibilities to reach the other side of the barrier if it is wide enough. In this case, it will be required a very strong wind in the proper direction.

4 Proposed Method for Urban Fire Spread Simulation The proposed algorithm to simulate the fire spread in an urban environment is an extension of the previously field definition. Thus, the information of the square cell is now enhanced with the urban data needed to simulate the fire spread. The proposed method introduces the concept of Building, which can be defined as a 2.5D representation of the real buildings using the approximation of the 2D projection of the building in the field and the number of floors. In order to fill the field with the information about the buildings, the proposed algorithm splits all the buildings, following the grid, in vertical cells called Building Units, which are split in other small units called Floor Unit.


4.1

139

Different Types of Buildings

Following the approach of Iwami et al. (2004), the proposed method differentiates three types of Floor Units: The Wooden Unit, Secure Building Unit (normal building), Shanty Unit (see Fig. 6). The main difference between the buildings types are their behaviour when they are exposed to fire. Secure Building Unit is the most secure type, being the most difficult type of building to get burned. By contrast, the Shanty Unit is the more dangerous one, too easy to get burnt, but with few quantity of combustible. The Wooden Unit is an intermediate structure, with the flash over effect, i.e., when a determined heat level is reached, the wooden structure itself is burnt very quickly, producing a blast which will increase the heat and the fire intensity.

4.2

The Fire Evolution in a Floor Unit

The evolution of the fire in a Floor Unit is characterized by four states. At the beginning of the fire, the ignition moment, the Floor Unit is in the state FloorSafe. When the fire begins to evolve in the unit, we are in the state FireOnlyInterior. This state is characterized by flames rising only from the openings, e.g., windows). Then, when the fire takes more power and begins to rise from the roof, we reach the state FireWholeFloor. At this moment, the fire is burning the whole Floor Unit. The last step is the third state characterized by a burned Floor Unit. The evolution of the fire depends on the type of the Floor Unit. In fact, a Secure Building Unit will have a fire evolution from state FloorSafe to FireOnlyInterior and then, to the final state FloorBurnt. The state sequence for the Wooden Unit is FloorSafe, FireOnlyInterior, FireWholeFloor and finally, FloorBurnt. The sequence of the Shanty Unit is FloorSafe, FireWholeFloor and FloorBurnt.

Fig. 6 Different types of Floor Unit’s. The Secure Unit is equivalent to a modern flat, very difficult to get burnt and isolated enough to reduce the probability to spread to contiguous floors. The Wooden Unit is normally used in building with up to four floors, very dangerous if they catch fire as they can literally explode in a blast. The Shanty Unit are normally referred to one floor low quality structures. They normally catch fire very easily, but the fire tends to extinguish itself very quickly as the available combustible is burnt almost instantaneously

140

4.3

A. Moreno et al.

Life Cycle of a Floor Unit

In order to create a specific behaviour where different entities (Floor Unit’s) will be able to interact between them, a life cycle has been set. When the fire starts, it only affects to the Floor Unit where it belongs. During this step, the state of the fire changes accordingly to the previously presented states. When the fire is burning inside the Floor Unit, heat is being released continuously, what could lead to spread the fire to the upper Floor Unit, but also to the neighbour buildings and Floor Unit’s (depending on the distance between them). Finally, the original Floor Unit will be totally burned, but the other ignited Floor Unit’s, contaminated by radiation will follow at their own the same life cycle.

4.4

Heat Released and State Transitions

The evolution of the fire in the Floor Unit’s is based on the released heat during the ignition (Babrauskas and Peacock 1992). The evolution from one state to other depends on the measured changes of the heat. In the proposed method, the heat evolves proportionally to the square of the time as long as the maximum heat is not reached. The left side of the Fig. 7 represents the heat evolution for a Secure Building Unit and a Shanty Unit. The difference between both Floor Unit’s is the time scale for every part of the curve and the maximum released heat. The right side of the Fig. 7 represents the heat evolution for the Wooden Unit. The heat of the Floor Unit’s does not exceed a maximum value, which depends on the surface of the Floor Unit’s. The proposed method takes into account several geometric characteristic to calculate the approximated evolution of the heat inside a Floor Unit: (1) the amount of combustible is directly proportional to floor surface; (2) the size of the windows openings, which influences the strength of the combustion, since it is proportional to the summed air flux between the exterior and the interior of the floor; and (3) the height of the Floor Unit. In the case of the Wooden Unit, it has another heat level of importance (maxHeat/2) to simulate the flash over effect.

A

B

Fig. 7 Heat evolution in different Floor Unit’s. The left side corresponds to the Secure Building Unit, where the evolution of the fire is slower. The right side is the heat released of a Wooden Unit, where an intermediate ignition burst can be found. In both cases, the heat released reaches to a maximum, and when the available combustible is consumed and the fire stops, the heat starts to cease, reaching to zero in a given amount of time


141

The last phase of the fire evolution starts when 90% of the combustible is consumed, and therefore, the heat begins to decrease linearly

4.5

Urban Fire Spread Between Floor Unit: Radiation Effect

The algorithm simulates the radiation effect between different Floor Unit’s, allowing the fire to spread to adjacent floors of a building or the ignition of other surrounding buildings. The proposed algorithm implements three types of radiation effect, (1) between floors of the same building, (2) between two different buildings and (3) between a BuildingUnit Façade and an OutsideUnit Area. Inside a building, the fire goes generally up, trying to reach to the superior levels. But it is also possible to be spread towards lower floors. In order to allow this possibility, a small probability of spreading to lower levels (10%) is given to the fire. The instant when a Floor Unit spreads to another floor is also randomly calculated, but always when the released heat is at its maximum, avoiding the spread when the fire is not strong enough. The radiation effect is also considered when two buildings are not in direct contact. In our implementation of the radiation effect, given two Floor Units of different buildings, the fire can be spread between them if the linear distance is less than 12 m and the release heat is higher than the 80% of its maximum value. Another factor to take into account is the existing wind (speed and direction), which will increase or decrease the spread probability. An existing fire on a Floor Unit can spread to the outside field through the Floor Unit’s exterior openings, located in its façade. If the Floor Unit has no outside façade, the fire will never go out directly from that floor.

5 Fire Extinguisher Model The proposed methods allow trying to extinguish an ongoing Urban/forest fire. Its main purpose is to allow the firefighters to interact with the fire evolution and see the consequences of their decisions on the field. We have two approaches for this model, one for the outside fire simulation, and another for the building fire simulation.

5.1

Fire Extinction for an Outside Fire

This model is conceptually the antagonist of the fire spreading algorithm, where the Fire Power amount is replaced by Fire Extinguisher Agent amount (e.g., water).

142

A. Moreno et al.

When, a given quantity of water is thrown into a cell, the same quantity of Fire Power is decreased. Thus, as long as there is water in a cell, its Fire Power will be equal to zero and the fire will be stopped (changing to the state FireStopped). In order to simulate the evaporation (see Fig. 8) of the water when it is thrown in an active fire, we use a similar approach of the radiation effect. A burning cell can reduce the quantity of water of the surrounding cells, depending on the distance between them and the Fire Power of the cell. In consequence, if a cell is in the state FireStopped but it is still containing food (combustible), the fire in that area could be restarted thanks to an evaporation effect due to the surrounding burning areas. The presented model is a very simplified version of how the fire agents fight the fire, but it is also generic. This model fits very well to simulate the actions of the ground firefighter, with a given quantity of water being thrown locally, or the firefighting aircraft’s, which throw a big amount of an especial extinguisher agent in a big area but for a short period of time.

Fig. 8 Evaporation effect. A reduction of water on the field is shown in the picture sequence (from the up to the down). The yellow sticks correspond to the water quantity


5.2

143

Fire Extinction for a Building Fire

The main concept behind the simulation of the extinguishment of the fire is that a fire extinguisher agent (or for clarity, water) can be thrown inside the building through the openings of a Floor Unit. Depending on the power of the fire extinguisher jet, it will be able to reach only the targeted Floor Unit or maybe, some internal Floor Unit of a building. When the water reaches a burning Floor Unit, its heat is reduced accordingly to the amount of received water. The fire will be stopped when the heat is equal to zero, but, in a similar way to the forest model, the Floor Unit will remain some residual heat that could eventually trigger some fire in the neighbours. If the water is not enough to stop the fire, the water will be evaporated almost instantaneously, but the heat in the room will be lowered. If the water is not thrown constantly, the fire will continue its evolution after the evaporation of the water.

6 Implementation and Results A C++ implementation of the algorithm has been done to test the algorithm behaviour and outputs. The tests have been performed using a field of 2.25 ha, choosing a tile side size equals to 3 m, which is half the distance of possible spread of fire when there is no wind (Breton and Duthen 2008). To simplify the tests, the wind is uniform and constant in the entire field and during all the simulation.

6.1

Forest Fire Results

The simplest test is to run the fire simulator without any external factor which can influence the fire spread (wind, slope). The fire evolves producing a in a circular shape (see Fig. 9, top-left). The slope effect has been tested under two scenarios, one to test how the fire can evolve towards higher positions (see Fig. 9, top-right) and to test how the fire avoids go towards lower positions (see Fig. 9, bottom left and right). The radiation effect has been tested using the scenario shown in the Fig. 10. The burning trees (pure green squares) facilitate the fire crossing river effect. Once the fire has crossed the river, it continues spreading in a normal way. It is remarkable that the fire spreads to the other side of the river, but in a much lower speed. After some time, the fire becomes stronger and recovers its maximum intensity.

144

A. Moreno et al.

Fig. 9 Some results of the Fire Spreading algorithm. Grey level shows the height map of the field: white, low height cells; black, high height cells. (Top-left) no wind, no slope. (Top-right) slope, the fire goes up the hill. (Bottom) slope, the fire avoids enter the valley, surrounding it

Fig. 10 Forest fire crossing a river (blue cells). Green cell is a tree. Black/Orange cells correspond to the fire spread

6.2

Urban Fire Results

Just before testing the urban fire simulator using an urban model, some preliminary test was performed to single building in order to check if the simulated fire evolution corresponds with the expected output.


145

Plot 0

2400 2200 2000 1800 1600 1400 1200 1000 800 600 400 200 0 0

50

100

150

200

250

300

350

400

450

500

550

Fig. 11 Heat (measured in MW) evolution of a wooden unit burning. The unit burnt in approximately 10 min. The flash over effect is around 150 s

Fig. 12 Both images represent how the fire spread in a building. The fire origin is located in the red box with more height, and it has been spread to the neighbour floors (small red box). The height of the red box represents visually the intensity of fire in that floor

This preliminary test (Fig. 11) simulates the evolution of a fire in a Wooden Floor Unit of 9 m3. The unit burned in 600 s with the heat evolution shown in Fig. 11. Although the heat evolution curve (and especially the maximum heat) depends on the type, height and openings size of the Floor Unit, the simulation behaves in the expected way, including the burst when the released heat reached at its half value. The relationships between buildings are tested by using a sample field, with buildings and forest areas, which will be used to test the radiation effect. If a fire is started in a building, if is not stopped, it will eventually spread. When the heat in a floor reaches a certain level, the other surrounding Floor Unit’s of the same level start to burn. In the Fig. 12, we can see two examples of how the fire spread to other floors.

146

A. Moreno et al.

Fig. 13 Radiation effect between near buildings (left) and the building and the surrounding vegetation (right). The fire origin is located in the big red box

Fig. 14 Capture of the prototype of the forest fire simulator, implementing the proposed method for forest fire spread. The place shown in the picture is an existing zone in the north of Spain

In a similar way, the radiation effect can occur between different buildings or a building and the near forest areas (see Fig. 13). In these cases, some wind must be present in order to ease the probability of fire spread.

7 Conclusions and Future Work In this paper, a pedagogical real-time fire simulation algorithm has been presented. Its main purpose is to be integrated into interactive virtual simulations where firefighter and managers can train their skills.


147

Although the forest fire spreading is a very complex phenomenon, we tried to simulate the most common characteristics of its behaviour by simplifying the model. Only two main physical variables have been used in the algorithm: direction of the wind, and the slope. The same approach has been followed for the urban fire simulation keeping only the heat parameter. The results have shown that the evolution in time of the fire fits the overall behaviour of the fire under different conditions. Some of the used approximations could be improved in the future. We can think about better utilisation of the wind information. The fires modify constantly the local wind, as it is changing the humidity and temperature of the atmosphere. The integration of a low complexity implementation of this phenomenon into the fire simulation would increase the realism of the output. Also, a comprehensive comparison between different algorithms should be done. These tests will try to measure analytically the efficiency, the memory consumption and other related factors. Since the ultimate objective is to integrate the algorithms in a VR environment, a high quality rendering should be implemented. Some preliminary results are shown in Fig. 14. After implementing the whole setup, user tests will have to be done in order to validate the algorithm output. Our preliminary results implemented in a 3D real-time application are based on the OpenSceneGraph graphical API, where some testing 3D urban and forest scenarios were created and loaded. The utilisation of high quality rendering techniques to render the fire, the smoke and the terrain gives a realistic impression of the simulation. The utilisation of real 3D city models can reduce the complexity of the model creation task and enhance dramatically the outcome of the results. The CityGML (Gr€oger et al. 2008) standard can deal with all the currently required information by the algorithm, but also, it is also suitable for future needs. As the number of CityGML models grows, they could be used as the field of the urban and forest fire simulation method presented in this work. For exchange and storage of virtual 3D City Models refer to the CityGML homepage (http://www.citygml.org/, last accessed [01-2011]). Acknowledgements This work was carried out in the context of project SIGEM, funded by the Spanish Industry Ministry through its Avanza IþD Programme and it was supported by COST Action TU0801 “Semantic Enrichment of 3D City Models for Sustainable Urban Development”.

References Achtemeier, G. L., 2003. An Application of Stephen Wolfram’s “New Kind of Science” to Fire Spread Modeling, USDA Forest Service, Athens. Babrauskas, V., and Peacock, R. D., 1992. Heat Release Rate: The Single Most Important Variable in Fire Hazard. Fire Safety J. 18(3):255–272. Breton, T., and Duthen, Y., 2008. Les simulations de propagation de feu en milieu urbain, http:// hal.archives-ouvertes.fr/docs/00/28/79/87/PDF/Les_simulations_de_propagation_de_feu_en_ milieu_urbain.pdf, last accessed [01–2011].

148

A. Moreno et al.

Dumond, Y., 2008. Forest fire growth modelling with geographical information fusion. 11th International Conference on Information Fusion, June 30–July 3, 2008, Page(s): 1–6. Finney, M. A., 1998. FARSITE: Fire area simulator-model development and evaluation, Res. Pap. RMRS-RP-4. USDA Forest Service, Rocky Mountain Research Station, Ogden, UT. 47 p. Forest Fires Website, 2010. Statistics on Forest Fires, http://library.thinkquest.org/07aug/01254/ statistics.html. Last time visited: 15/05/2010. Gr€oger, G., Kolbe, T. H., Czerwinski, A., and Nagel, C., 2008. OpenGIS City Geography Markup Language (CityGML) Encoding Standard. Version: 1.0.0, OGC Doc. No. 08-007r1, Open Geospatial Consortium. Hamada, M., 1951. On Fire Spreading Velocity in Disasters, Sagami Shobo, Tokyo (in Japanese). Iwami, T., Ohmiya, Y., Hayashi, Kagiya, K., Takahashi, W., and Naruse, T., 2004. Simulation of City Fire. Fire Sci. Technol. 23(2):132–140. Lee, S., Davidson, R., Ohnishi, N., and Scawthorn, C., 2008. Fire Following Earthquake – Reviewing the State-of-the-Art of Modeling. Earthquake Spectra. 24(4):1–35. Linn, R., Reisner, J., Colman, J. J., and Winterkamp, J., 2002. Studying Wildfire Behavior Using FIRETEC. Int. J. Wildl. Fire. 11:233–246. Ohgai, A., Gohnai, Y., Ikaruga, S., Murakami, M., and Watanabe, K., 2005. Cellular Automata Modeling For Fire Spreading As a Tool to Aid Community-Based Planning for Disaster Mitigation. Book, Recent Advances in Design and Decision Support Systems in Architecture and Urban Planning, Kluwer, Dordrecht, The Netherlands, 193–209. Richards, G. D., and Bryce, R. W., 1995. A Computer Algorithm for Simulating the Spread of Wildland Fire Perimeters for Heterogeneous Fuel and Meteorological Conditions. Int. J. Wildl. Fire. 5(2):73–80. Rocky Mountain website, 2010. Aircraft Rescue and Fire Fighting Basic Training, http://www. rmestc.com/. Last time visited: 15/01/2010. Rothermel, R.C., 1972. A Mathematical model for Predicting Fire Spread in Wildland Fuels, Research Paper. U.S. Department of Agriculture, Forest Service, Intermountain Forest and Range. Scawthorn, C., Eidinger, J., and Schiff, A., 2005. Fire Following Earthquake, Technical Council on Lifeline Earthquake Engineering Monograph No. 26, American Society of Civil Engineers, Reston. Schmuck, G., 2006. Forest Fires in Europe, Report No. 7/2006, European Commission, Joint Research Centre Institute for Environment and Sustainability, Italy. Tanaka, T., and Himoto, K., 2006. Physics-Based Model of Urban Fire Spread and Mitigation of Post-earthquake Fire Risk in Historic Cities. Abstracts for Annuals, Disaster Prevention Research Institute, Kyoto University. 49(B):1–5. Weise, D. R., Biging, G. S., 1996. Effects of Wind Velocity and Slope on Flame Properties. Can. J. For. Res. 26:1849–1858.

3D Cadastre in the Province of Quebec: A First Experiment for the Construction of a Volumetric Representation Jacynthe Pouliot, Tania Roy, Guillaume Fouquet-Asselin, and Joanie Desgroseilliers

Abstract The current cadastral system in the province of Quebec is a graphical one in the sense that it presents the limits and the official measures of the property on a 2D digital map. To be able to represent superimposed properties like condominium, the Quebec cadastre uses “le cadastre vertical” that is a polygon with a number that refers to an external complementary plan (PC). This plan shows vertical profile of the properties and a detail draw of each floor (private and common parts). A single PC-number could refer to hundreds of lots and plans depending on the geometric complexity of the building. The understanding of the spatial arrangement of all superimposed properties contained in the PC file is a tricky mental exercise. To help users of the cadastre vertical, a semi-automatic procedure is proposed that enables the construction of a volumetric representation from the PC image file. In this specific constraint situation, the various data processing steps are described starting with the vectorization (from image to vectors), the 3D modeling (the construction of the volumetric representation) and finally the data exchange. The ins and outs of every data processing, the time and efforts required to achieve each step are discussed, and we conclude with remarks made by the end users about potential usages of such cadastral volumetric representation.

J. Pouliot (*), T. Roy, G. Fouquet-Asselin, and J. Desgroseilliers Geomatics Department, Laval University, 1055, avenue du Se´minaire, Que´bec, QC, Canada G1V 0A6 e-mail: [email protected]


149

150

J. Pouliot et al.

Fig. 1 Window from the Web interface Infolot that allows viewing cadastral parcels of the province of Quebec ( for privacy reason identification were hidden)

1 The Current Quebec Cadastral System In the province of Quebec, Canada, the property transactions are registered into the land registration system. This system includes a document called index of immovable’s that describes the transactions and a cadastral plan. The cadastral plan,1 as shown in Fig. 1, is a 2D graphical description of the limits and the size of the lot (the parcel) where each property has its own ID (unique lot number). The ministe`re des Ressources naturelles et de la Faune (MRNF)2 also called Foncier Que´bec is the official responsible for managing and maintaining the land registration infrastructure. We currently count around 3,500,000 properties in the province of Quebec and by March 2010 more than 2,500,000 parcels were renewed by the Cadastral Reform.3 Infolot4 is an online interface that allows everyone to view the cadastral plans and for registered clients to get copy of it. As part of the land registration system the Quebec cadastral plan plays important roles for the immatriculation, the representation of the properties and thus the management of right, responsibility, restriction (RRR) associated to it. It also serves to establish property taxes, to help urban planning and management tasks (e.g. public utilities) and could be used for the application of various laws and regulations. Therefore the cadastral plan has to be up-to-date, complete and non ambiguous. In most cases the Quebec 2D cadastral plan performs well to fully represent the limits of

1

See the official publication Code Civil du Que´bec (C.c.Q.), rules 3026 3027 3028 3028.1 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043. 2 http://www.mrnf.gouv.qc.ca/english/land/index.jsp 3 http://www.cadastre.mrnf.gouv.qc.ca/planification/bilan-travaux.asp 4 http://www.infolot.mrnf.gouv.qc.ca/

3D Cadastre in the Province of Quebec: A First Experiment for the Construction

151

Fig. 2 An example of the cadastral plan that refers to a complementary plan (PC)

the property including the lot number and its official measurements such as width, length, and surface. In some situations where superimposed properties like condominium exist, the classical 2D plan is not sufficient to fully represent the width, length and height of the properties. That is why the cadastral instructions of the MRNF (see MRNF 2003) include specific status called “cadastre vertical”. In brief the cadastre vertical allows the representation of the vertical limits of the property, in the specific situation of multiple uses of space. To be able to distinguish this third dimension, the cadastral plan will refer to a complementary plan5 (PC). Figure 2 presents an example of the cadastral plan referring to a specific PC (indicated by the PC-number). We can observe that there is no lot number and no official measurements associated to this polygon. The PC is available on demand, usually stored as a pdf file. The PC are created by land surveyors (private firms) who did the job of measuring the properties and its drawing. The private firms usually stored the original PC as a CAD file but this one is not available for others users and stays belongings of the private firm. The end users of the PC only have access to the pdf file (an image of the draws). As shown in Fig. 3, this file contains scans of the original CAD file. The PC contains a localization view of the building linked to the boundaries of the common lot, a 2D draw of each floor containing private lots and a vertical profile of each lot. Only private lots have official measurements such as length, width, and height. Usually, the ground, basements, foundations, main walls of the building are common parts and represented by only one lot number without official measurements except for the localization view for the external ground.

5

Also call supplementary plan.

152

J. Pouliot et al.

Fig. 3 Example of images extracted from the PC file. (a)Vertical Profile of the superimposed properties, (b) draw of the second floor

2 A First Experiment to Produce a Volumetric Representation Taking into consideration this current situation of the cadastral system in the province of Quebec, few elements can be stated. When superimposed properties exist, the representation of the third dimension is available, but the current solution is not fully integrated into the cadastral system. We need to obtain the PC file and manage it independently of the cadastral plan which restricts the interaction and potential analysis and querying. This file is created by a scanning process that limits the overall quality (depending on the scanning resolution) and we thus lose the access to the original document (the CAD file). A single PC-number could refers to hundred of plans depending on the geometric complexity of the properties. From this pdf file, the understanding of the spatial arrangement of every superimposed


153

property, a must consideration associated to the cadastre tasks, is a tricky mental exercise. Experience is required! We thus believe that a volumetric representation for the cadastre would greatly help understanding the space arrangement of lots and by the fact improve the management of the RRR associated to them. Several previous works on 3D cadastre can be found in the literature and all of them present advantages of using such a volumetric representation. Stoter’s works are certainly the main comprehensive and exhaustive source of information (Stoter and Ploeger 2003; Stoter 2004; Stoter et al. 2004; Stoter and van Oosterom 2006). This literature review was of great help for us to better understand the issues and to be able to propose a more adequate solution (for more detail see Pouliot et al. 2009). In order to demonstrate the helpfulness of a volumetric representation of the cadastre, an undergraduate geomatics engineering student’s project was designed (a 4 months project). From the previous scenario (i.e. having access from Infolot to the 2D cadastral plans and a PC scan file), the objective of this project was to produce as automatically as possible a volumetric representation of each individual 3D lots. We identified several categories of end users that have various levels of knowledge and expertises related to spatial data management and cadastral applications and use the cadastre for a range of tasks. Foncier Que´bec, the official responsible for managing and maintaining the cadastral plans was one of them, to whom were added notaries, land-surveyors, urban planners, real estate agents and even the general public (the owner itself). Thanks to various discussions with the end users few requirements and constraints associated to this task were identified. First the time required to build the volumetric representation of the lot was one of the main constraint identified. We estimated that between 10 and 15 min/PC file would be the maximum acceptable delay. The second aspect was related to the automation; the procedure will have to be as automatic as possible. Less than 15 clicks or manual interventions were our initial target. As mentioned, the end-users could have various levels of knowledge and expertises related to spatial data management and this aspect will have to be taken into account. From a more technical point of view, the volumetric representation of the lot has to keep its lot number and if possible gives access to the official measurements. The 3D representation will have to be a close solid element with a precision in X, Y and Z as similar as the cadastral plan specifications (0.3 mm graphical tolerance depending on the scale of representation). No specific requirements were identified concerning the geometric modeling approach; it could be Constructive Solid Geometry (CSG), Boundary representation (B-Rep), extrusion, voxels, tetrahedron representation as long as the procedure respects the construction time and automation constraints.

3 The Workflow to Build 3D Lots The proposed workflow to create volumetric representation of superimposed properties from PC files is organized around three general phases (vectorization, 3D modeling and data exchange). In order to automate as much as possible the data

154

J. Pouliot et al.

processing, some programs or interface were written where the user is prompted for information required by the procedure. The end solution is not fully automated and we need to use a combination of automatic and manual interventions to obtain the desired result. By now, the procedure was setup on one PC file (or one PDF file) and tested on another PC. Both PC refer to four to six cadastral lots that could be owned by different peoples where each PC contains one ground parcel, one building with three to five apartments and common parts (walls, stairs, etc.). The next sections present the workflow and the discussion puts a specific emphasis on the time required and the automation aspects of the procedure since it was the two main constraints identified by the end users.

4 Vectorization The main goal of this phase is to obtain a vector file in which the localization view and each lot extracted from the PC image file will be georeferenced as a close-shape polyline and ready for 3D modeling. The vectorization has to produce clean vectors with content as similar as possible to the original data. Completeness, topology and spatial accuracy were the criteria used to decide if the output was acceptable. In order to take advantage of existing solutions, we selected and tested several software for vectorizing images such as RasterDesign (Autodesk), RasterVect, VHPhydrid CAD, WinTopo and character recognize software (OCR) such as AbbyyFineReader, SimpleOCR, OmniPage and RasterDesign. Table 1 shows the various data processing steps and the time required to execute the vectorization. The pdf file is first converted into a tiff file that allows the use of image filtering techniques to enhance the render of the lines before executing the vectorization. Some criteria have to be identified to control the quality of the resulting image. For instance this process will have to produce no isolated pixel, no hole in line, no double line appearing, pixels representing a line are continuous. Once the image filtering is satisfying, some parameters were set for the vectorization itself. The quality control was driven by the minimal length of line, the spatial precision and the straight lines recognition precision. Data cleaning task is

Table 1 Data processing for the vectorization of one PC file

Data processing

Convert pdf to tiff Image enhancement Parameterization of vectorization Convert image to vectors Data cleaning Georeferencing the localization plan

Number of manual intervention 1 1 1

Time required (min) 1 1 0.5

1 48–115 4

2 10–25 5


155

crucial and several iterative tests were made to find the appropriate set of parameters to help the users. The final step consists in georeferencing the PC; this was done by an affine transformation with four control points entered by the user on the localization plan. In overall, the vectorization takes between 20 and 30 min for one PC. The main sources of variation come from the data cleaning process. For example, the presence of common walls, doors, stairs, or a lot distributed on two floors augment the complexity of the task and thus often require manual intervention. The time required in our procedure is obviously higher than the original target. On the other hand, 93% of the lines (their coordinates) were inside the spatial accuracy tolerance, meaning that the vectorization process does not alter too much the shape of the lines. During this process, several issues have to be mentioned. First the optical characters recognition (OCR) task is a complex procedure when it is not applied to regular text manuscript. For instance as shown by Fig. 4, annotations (or official measures) have various orientations and sizes (scale) in the map. In some cases they were placed very close to the line; so close that the software mixes up the text with the line. And finally, the OCR does not record the position of the text, loosing as a result the possible automatic matching between the annotation and the corresponding line. To be effective, an OCR procedure would have to ask the user to manually specify one annotation at a time that has to be converted. This last data process was tested with the solution proposed by Autocad+Raster Design. Even if it was working correctly, it was too much time consuming and this solution was finally discarded. The proposed solution is consequently not recording the official measurements, only the lot number is manually entered. Another problem is the presence of arrows on the plan (as shown on Fig. 4). During line extraction, some arrows were merged with the cadastral lines, which is obviously not acceptable. Furthermore, some lines are not continuous (e.g. dot lines). Data processing can be used to close these lines but some of them were corresponding to real hole (like door), the distinction between right and wrong holes was thus difficult to perform automatically. Manual data cleaning was finally required to complete the process.

663

971

0,14

0,14

2

1,15

h.: 2.06

Fig. 4 Example of text annotations in a PC file having various orientation and size

156

J. Pouliot et al.

5 3D Modeling The phase of 3D modeling consists of converting the georeferenced close-shape polylines into individual volumetric elements (one volumetric primitive will correspond to one cadastral lot). Some tests and a literature review were conducted to evaluate 3D geometric modeling approaches or existing standards (among others BIM, CityGml, architectural, engineering) to find which one was more appropriate to our context (Be´dard 2006; Benner et al. 2005; Janssens-Coron et al. 2009; Kolbe et al. 2005; Ledoux and Meijers 2009; Lee et al. 2008; Marsh 2004; Shen et al. 2007). But finally to keep the process short, easy and satisfying the previous requirements/constraints we decided to use simple extrusion and CSG techniques to produce the 3D lot. We ended with a volume (3D Lot) bounded by flat faces. This approach was easy and fast to implement and several CAD software offer algorithms to build the volumetric element. The topological consistency check during 3D modeling was limited at having closed volume for one cadastral lot in one PC; a sharing wall between two lots will be thus duplicated. The CAD software MicroStation from Bentley was used since it proposes all the required functionalities for 3D geometric modeling and it is easy to add batch processing, user’s interface and quality control program. Table 2 shows the various data processing steps and the time required to execute the 3D modeling. First, the right elevation and horizontal positioning is assigned to the lots. The next step is to extrude each lot (its boundary) in order to produce a simple block solid element (the 3D lot). The final step seeks at finalizing the 3D lot. It could include various data process depending on the content of the PC. For example when bearing walls are present and because they do not belong to the private parts, they have to be individually extruded and subtracted from the initial 3D lot. The same process has to be done for some doors that do not belong to the private parts. In overall, the 3D modeling takes about 10 min and requires at least 50 interventions for one PC. The main source of variation come from the finalization step that depend on the PC content (e.g. presence or not of bearing walls, doors, stairs, or lot distributed on two floors). Our experiment and results were obviously restricted by the number and the variety of PC tested. The following pictures show the volumetric representation of one PC (or one condominium) (Fig. 5). Table 2 Data processing for the 3D modeling of one PC file Data processing Number of manual intervention Extraction of each lot and gathering 6–10 its height Co-registration with the localization 9–16 plan and add the scale factor Extrusion of cadastral lots 5–18 Finalizing the 3D lots (subtracting 31 the bearing walls and the doors)

Time required (min) 2–3 1 1–2 5


157

Fig. 5 (a) One condominium with three apartments, (b) The corresponding volumetric representation accessible in the Google Earth view

158

J. Pouliot et al.

6 Data Exchange In order to give a better access to 3D models by the end-users, few tests of data import/export were performed. For this task, some priorities were established such as having access to a free 3D viewer, being able to make measurements and extracting 3D slices, being able to integrate the 3D model with other existing 3D models such as those proposed by the city of Quebec and Google Earth and finally keeping the integrity of the 3D model (in terms of coordinates and the cadastral lot ID number). Different solutions such as ArcScene (ESRI), FME Data Inspector (SafeSoftware), LandXplorer (Autodesk), Myriad 3D Reader, 3D PDF (Adobe) and Google Earth were tested. Each one has its own characteristics and even if several 3D formats are available for import/export, the result has to be carefully verified. For instance, few problems of compatibility between Microstation SmartSolid/line and vrml (Myriad 3D) and LandXplorer (the geometry was altered) were observed. The superposition of external 3D models was easily realized with Arcscene and Google Earth. Few troubles in the management of the Z coordinate were also detected; the datum has to be carefully adjusted. Google Earth importation has to take into consideration the conversion of the MTM coordinates system to latitude/longitude. Manual conversion can be done or COLLADA interchange file format can be used to achieve this task. 3D PDF file was quite easy to create and easily accessible by anyone. It offers few interesting functionalities; we can rotate, zoom, measure, having access to the hierarchical structure of the 3D model (each level correspond to a 3D lot). It is however not possible without specific data processing to integrate the 3D PDF or the vrml files with others 3D models.

7 The Validation with the End Users To validate the fitness for use of our procedure and the 3D model itself, several meetings were organized with the end users. As mentioned, different categories of end users and usages were identified that help us to specify the project targets. We interviewed Foncier Que´bec, two notaries, the city of Que´bec (a first group working in architecture and patrimony and a second group responsible of mapping and surveying) and four land surveyors. The users interviewed were for the most expert or having good knowledge about cadastral data. At each meeting, the framework was presented and discussions about the time needed, the automation, the knowledge and expertise required and the level of detail (LoD) for the content of the 3D representation were conducted. About the volumetric representation itself, various degrees of interest were reported depending on the category of end users. Foncier Que´bec would anticipate using it as an effective mechanism for data validation and quality control. Data quality control of PCs currently requires several efforts and specialized resources and this task could directly profit of having integrated 3D lots. The notaries would


159

be interested in having 3D model but mostly for the representation of complex situation of superimposed properties. For the one experimented (i.e. a medium size condominium) they do not perceive at a first glance the interest of having a volumetric representation since the comprehension of the spatial arrangement of the properties is not enough complicated. The municipal authorities (both groups) were very interested to get a 3D cadastre representation. A 3D model would give them the opportunity to have a complete overview of the geometry of the property and thus improve other systems like taxation, water supply and sewer systems. They would have it integrated as a 3D database from which they could query the 3D lots and add several attributes required for accomplishing their responsibilities. They would apply several 3D spatial analyses like 3D buffer to assess whether or not a property meet certain regulations (e.g. historical protection area). Finally, the land surveyors agreed about the importance of having such 3D representation but not necessary for their own uses. They recognize a more attractive usages for others specialists such as notaries and real estate agents and even the general public. They also mentioned the importance of having a volumetric representation in the case of “proprie´te´ superficiaire”6 where the owner of the land is not the same as the owner of the building. The current Quebec cadastral system does not always show this kind of situation. About the procedure for building a volumetric representation, all the users recognized the necessity of having an automated process to build the volumetric representation since the knowledge about 3D modeling is generally low if not absente. According to the end users, the automation and the time constraints were the most important aspects to be addressed. The initial target for the time delay was about 10–15 min/PC. We thus had some failures on this side because the complete procedure took in overall around 30–40 min to process one PC. The experiments nevertheless highlight the fact that the vectorization phase which is a long and problematic step could easily be withdrew if the original CAD file instead of the pdf file was available to the end-users. As mentioned the original CAD file still belong to the private firms of land surveying. If the original CAD file was available, the proposed procedure would then require less than 10 min. But this result was obtained for one PC of simple condominiums with four to six lots and no official measures were recorded. The experience also clearly demonstrated the importance of taking into account the complexity of the geometry of the buildings and the level of details (LoD) required. Four possible LoD were identified ranging from generalized lines without any official measures to more detail lines with official cadastral information. The LoD concept was important since it helps us to find a more appropriate solution associated to time and automation constraints. For further development, this aspect of LoD will have to be tackled and explained adequately to the end users. About the knowledge and expertise required, Foncier Que´bec, and both groups at the city of Quebec were receptive to use specialized software as proposed since

6

This French technical term could be translated by “right of superficies”.

160

J. Pouliot et al.

they already work with this kind of products. But some others like the notaries are not really attracted to learn new software, what is understandable. Finally, the land surveyors were not really interested into the procedure since they build themselves and possess the CAD file and could then having access to their own techniques (this is mostly true for the vectorization phase). They nevertheless realized that instead of preparing CAD file with profiles and 2D draws of each floor they could change their own procedure and build by themselves 3D models. This new opportunity could consequently require more expertise/training in 3D geometric modeling.

8 Conclusion The notion of 3D cadastre is not new but effective implementation of an integrated volumetric representation in national cadastral systems is often not available. Few countries have interesting solutions for 3D cadastre like Australia, Netherland or Japan, and the province of Quebec is certainly well positioned with its concept of cadastre vertical. We presented in this paper a workflow and outlined restrictions and obstacles associated to it that allows the construction of a volumetric representation from 2D plans (pdf files) containing vertical information. The discussions with several end users reveal that the proposal is worthy and could satisfy various needs, but not at any price. The time constraint was one of the main limits discussed in our paper. The main sources of incertitude and human efforts certainly come from the data cleaning steps and the finalization of the 3D lots to be able to respect geometric and topologic constraints. These variations are mainly due to the existence of various shapes of condominium, the presence or not of common parts such as walls, doors, stairs, and to MRNF specifications that evolved. The proposed procedure has nevertheless the advantages of being simple, concrete, and fully integrated with the current cadastral system of the province of Que´bec. It could certainly help other organizations interested to similar questions of having a specific workflow that enable 3D volumetric representation from current cadastral system. This work was completed in 4 months and did not allow us to achieve several important issues. One of the main restrictions is the limited number of PC under which the experiments were done. The procedure was tested on condominiums having a small number of apartments and regular geometry, which are somehow representative of the PCs found in the cadastre du Que´bec. Nevertheless, in more complex situations where more than 20 owners could share the space over than 60 legal lots and where the building has irregular geometric shapes the proposed procedure could easily fail to respect the initial constraints. Also, condominium is not the only situation where we can find multiple usage of space. Subways, tunnels, underground parking are other kinds of situation that need to be investigated. We did not have time to interview real estate agents and owners; they could certainly give us another point of view about our proposal and the 3D model itself. We did not achieve any test about the importance of having texture or others kinds


161

of information (semantic or graphical for example) that could improve the use of 3D cadastral representation. At the beginning of this project a database (DB) solution was identified as a valuable approach. We did not end with a 3D DB but at least the 3D model could be integrated into a DB system. Besides the 3D topology consistency is somehow weak and only checked for one lot; this aspect should be taken into account because it could restrict the 3D analysis capabilities and the 3D model quality itself. Finally, for this first experiment the problem of 3D cadastre was investigated from a point of view of engineers (3D model construction). We did not explore the eventual impacts on laws and regulations of having volumetric representations of the property or any questions of ethic related to the production or the use of such 3D model (see www.3dok.org). Acknowledgements This project was funded by the Natural Sciences and Engineering Research Council of Canada, as a personal discovery grant. A special thank to E. Janssens-Coron and L. Hashemi Beni for great help in the progress of this work. We would also thank all end-users met during this project, from MRNF Foncier Que´bec M. Morneau and L.-A. Desbiens, as landsurveyors M. Gervais, F. Roy, B. Beaulieu and N. Masse´, as notaries F. Brochu and B. Roy, and from the city of Quebec B. Fiset, M.-A. Bluteau, B. Couture, A. Tremblay, A. Naud.

References Be´dard K (2006) La construction de mode`les geólogiques 3D a` l’e`re de la normalisation. Me´moire de maıˆtrise, Sciences geómatiques, Universite´ Laval Benner J, Geiger A, Leinemann K (2005) Flexible generation of semantic 3D building models. In: First international ISPRS/EuroSDR/DGPF-Workshop on next generation 3D city models, vol. 49, EuroSDR Publication, Bonn, Germany, pp. 17–22 Janssens-Coron E, Pouliot J, Moulin B, Rivera A (2009) An experimentation of expert systems applied to 3D geological models construction. In: Neutens, T. and DeMaeyer, P. (eds.) Lecture notes in geoinformation and cartography: Developments on 3D geo-information sciences, Springer, Heidelberg, Germany, pp. 71–91 Kolbe TH, Gr€oger G, Pl€ umer L (2005) CityGML – Interoperable access to 3D city models. In: Oosterom, P., Zlatanova, S., Fendel, E.M. (eds.) Geomatics solutions for disaster management: Geo-information for disaster and management, Springer, New York, pp. 883–899 Ledoux H, Meijers M (2009) Extruding building footprints to create topologically consistent 3D city models. In: Krek, A., Rumor, M., Zlatanova, Z., Fendel, E. (eds.) Urban and regional data management – UDMS Annual 2009, Taylor & Francis Group, London, pp. 39–48 Lee S, Feng D, Gooch B (2008) Automatic construction of 3D models from architectural line drawings. In: The Symposium on interactive 3D graphics and games, ACM, Redwood City, pp. 123–130 Marsh D (2004) Applied geometry for computer graphics and CAD. Springer, London MRNF (2003) Instructions pour la pre´sentation des documents cadastraux relatifs a` la mise a` jour du cadastre du Que´bec. Version 2.0, Fe´vrier 2003, document re´dige´ par le ministe`re des Ressources naturelles du Que´bec (MRNF), Direction de l’enregistrement cadastral, Gouvernement du Que´bec Pouliot J, Hashemi Beni L, Roy F, Gervais M, Brochu F (2009) La 3e dimension et sa pertinence pour des applications cadastrales. In : Confe´rence de l’association canadienne des sciences geómatiques, Montreál, Canada Shen Z, Issa RRA, Gu L (2007) Semantic 3D CAD and its applications in construction industry – an outlook of construction data visualization. Visual, LNCS 4781: 461–467

162

J. Pouliot et al.

Stoter JE (2004) 3D Cadastre. Ph.D. thesis, Delft University of Technology Stoter JE, Ploeger HD (2003) Property in 3D-registration of multiple use of space: Current practice in Holland and the need for a 3D cadastre. Computers, Environment and Urban Systems 27: 553–570 Stoter JE, van Oosterom PJM (2006) 3D Cadastre in an international context: Legal, organizational and technological aspects. Taylor & Francis, Boca Raton Stoter JE, van Oosterom PJM, Ploeger HD, Aalders H (2004) Conceptual 3D cadastral model applied in several countries. In: FIG Working Week, Athens, Greece, May 22–27

3D Modeling for Mobile Augmented Reality in Unprepared Environment Vincent Thomas, Sylvie Daniel, and Jacynthe Pouliot

Abstract The emergence of powerful mobile smartphones, with embedded components (camera, GPS, accelerometers, digital compass), triggered a lot of interest in the mobile augmented reality (AR) community and new AR applications relying on these devices are beginning to reach the general public. In order to achieve a rich augmentation in terms of immersion and interactions, these mobile AR applications generally require a 3D model of the real environment to provide accurate positioning or to manage occlusions. However, the availability of these 3D models based on real spatial data is limited, restraining the capacity of these applications to be used anywhere, anytime. To overcome such limits, we developed a framework dedicated to the fast and easy production of 3D models. The proposed solution has been designed for the specific context of mobile augmented reality applications in unprepared environment and tested on iPhone.

1 Introduction For some years, smartphones made a breakthrough in the telecom market. According to the information technology research and advisory company Gartner, Inc. “Smartphones continued to represent the fastest-growing segment of the mobiledevices market” (Gartner, Inc.). These mobile devices are powerful, small and involve several positioning, orientation and optical components: Global Positioning System (GPS) receiver, digital compass, accelerometers and camera. These components, in addition to the smartphone mobility, make this device a valuable platform for the development of unique and innovative activities or locationbased services such as mobile augmented reality applications. Augmented reality (AR) is the enrichment of the reality with virtual elements. These synthetic objects are projected in a live video stream as if they were part of

V. Thomas (*), S. Daniel, and J. Pouliot De´partement des sciences geómatiques, Universite´ Laval, 1055, avenue du Se´minaire, Que´bec, QC, Canada G1V 0A6 e-mail: [email protected]; [email protected]; [email protected]


163

164

V. Thomas et al.

the reality. Recently, some mobile augmented reality applications based on smartphones have been released like Layar (http://www.layar.com) or Wikitude (http:// www.mobilizy.com). These applications are world browsers that enable a user to visualize virtual layers of 2D–3D information superimposed to the camera video stream. A tourist in New York City can see annotations with names and information about the skyscrapers in the camera field of view for instance. One of mobile augmented reality advantages is to clearly link information with real object as seen by a person on the spot. The principle on which these applications rely does not involve information about the user surroundings when rendering the digital graphics. As a result, occlusions are not managed. Thus, a whole 3D graphic element will be rendered even if part of it is occluded by a real building for instance. To manage occlusions, it is recommended to integrate information or knowledge about the environment. Having a 3D model of the environment where the augmentation takes place offers opportunities for richer interactions and higher levels of immersion. More specifically, 3D models about objects in the user field of view can be used to merge more accurately the computer-generated elements with the reality, providing the user with a better sense of presence of these elements in the real world. However, these 3D models of the environment are not always accessible, therefore limiting the capabilities to deploy realistic augmented reality solutions anywhere, anytime. Mobile augmented reality is anticipated as the next social communication and information hub. Following the current trends of social networking and locationbased applications, citizen can already be foreseen as the main users and contributors of such technology. The real-world information used for mobile AR will be usergenerated, as applications like Wikitude or Yelp (http://www.yelp.com) are already demonstrating. Within this context, tools to create contents for mobile AR need to be simple and intuitive to enable the user to augment his environment with information that matters to him and to deploy such applications wherever he chooses to. This paper will present a new citizen based 3D modeling solution to enable mobile AR applications anywhere, anytime. The context and principles related to mobile AR will be first presented. A brief review of current mobile AR work will be provided too. Then, the proposed 3D modeling approach will be described as well as the prototype implementing such an approach. Finally, results of tests conducted to assess the prototype performances will be provided and discussed before concluding.

2 Mobile Augmented Reality Operational Constraints The concept of an augmented reality environment is often presented using the Reality-Virtuality continuum of Milgram et al. (1994) (cf. Fig. 1). Reality, namely what is directly perceived by a person or indirectly by a camera, stands at the left extremity of the continuum. A virtual environment, which is completely synthetic, stands at the right end of the continuum. Between both extremities are all the mixed reality environments, which merge computer generated graphics and real elements.

3D Modeling for Mobile Augmented Reality in Unprepared Environment

165

Fig. 1 Reality-virtuality continuum (Milgram et al. 1994)

Augmented reality, which is situated next to reality, is a real environment augmented by virtual elements. If the environment is predominantly synthetic with some real elements included, this context is related to augmented virtuality. Augmented reality approaches are relevant to many contexts, like, for instance, the maintenance of mechanical engine (Henderson et al. 2007), or teaching and learning 3D contents (Shelton et al. 2002). Augmented reality offers also an exceptional potential for gaming (Lindt et al. 2007). An augmented reality application is said to be mobile if the user is his own avatar and his position in the synthetic world follows his displacements in the real environment (Broll et al. 2008). Those types of applications can display various levels of augmentation. Mobile applications with low augmentation (i.e. “weakly augmented”) consist generally in displaying multimedia elements triggered by the physical proximity of the user to a virtual point of interest (POI). A POI is simply information, such as image, video or text, linked to a specific location at the surface of the Earth. Games described in Squire and Mingfong (2007) or Klopfer and Squire (2007) are good examples of such applications. Increasing the level of augmentation will increase the level of immersivity and interactivity of the application and, as a result, the capability to challenge the user interest. Applications with such level of augmentation are said to be “strongly augmented”. Literature review shows that very few strongly augmented applications have been developed so far. Most of them used head mounted display, a laptop, a GPS receiver and an inertial system as their hardware configuration. However, these devices were rarely integrated and definitely not ergonomic, limiting their deployment and adoption by user community. With the emergence of lightweight and powerful smartphones equipped with GPS receiver, digital compass, camera and accelerometers, there are new opportunities to implement mobile AR applications using these platforms and to bring mobile AR outside the research community. Layar 3D application (Layar 2009) is already a good example. Having specified the various declensions of mixed and augmented reality, we will now address the requirements of strongly augmented solutions. They will be the focus in the following paragraphs and sections of the paper. According to Azuma definition of augmented reality (Azuma et al. 1997), computer-generated graphics and reality should be perfectly co-registered in real time to provide the user with a strong augmentation of the reality. This requirement allows maintaining the user feeling of presence and immersion at all time. Good co-registration of virtual elements in the real world implies to precisely determine

166

V. Thomas et al.

the user position and orientation and to tackle occlusion issues. These positioning and occlusion issues are generally resolved using 3D models of the environment (Schmalstieg et al. 2007; Ohlenburg et al. 2007). To correctly manage the occlusions, the 3D model of the real environment has to be first integrated in the same virtual world as the graphic elements. Then, during the rendering of the scene, the model of the real environment is drawn in a transparent manner so the background (i.e. the camera live video stream) could be shown in the foreground in lieu of the occluded parts of the 3D virtual graphic elements. To retrieve a more accurate position and orientation of the user, the 3D model can be projected in the current video frame according to the previous camera pose computation as proposed by Reitmayr and Drummond (2006). Using computer vision algorithms, the 3D model projection is compared to the model features in the current frame of the video stream. The detected differences will provide information about the camera movements. They will complement the pose and orientation measurements provided by the hardware components (i.e. digital compass, accelerometers, GPS receiver). As it has been underlined above, a 3D model of the environment is required in order to build mobile AR applications with rich interactions and high immersivity. Various sources for 3D models currently exist. Virtual globes such as Google Earth or Bing Maps contain a handful of 3D models. However, it is not possible to retrieve them seamlessly depending on the user’s location. Actually, the only way to access a single 3D model from Google Earth is to manually download it from the 3D warehouse website (http://www.sketchup.google.com/3dwarehouse/). Open 3D servers exist like OpenStreetMap-3D (Neubauer et al. 2009), but they offer limited coverage and they do not seem to have a strong community leverage. On the other hand, 2D data is more publicly accessible and the missing elevation data could be derived from other sources (in situ observations, building’s altitude by adding the number of stories to the ground height). This data process still needs to be explored in the context of mobile augmented reality modeling. Taking into account that the availability for 3D models is currently limited, the constraint of having 3D models anywhere at anytime in mobile AR applications cannot be always satisfied. If a 3D model of the environment it is not available or cannot be retrieved, the environment is said to be unprepared for augmented reality. The proposed citizen based 3D modeling solution presented in this paper specifically addresses this context of unprepared environments allowing the user to capture on location the 3D model he needs. The next section describes the criteria such 3D modeling tool should fulfill to supply 3D models adapted to mobile AR application.

3 3D Modeling Solution Specifications for Unprepared Environment The context of mobile AR in unprepared environment focus on augmented reality anywhere, anytime. As mentioned in the introduction, the general public is one of the main targets of such application. Therefore, it can be assumed that mobile AR applications in unprepared environment should rely on limited external devices


167

beside the smartphone itself and should not require specific expertise in computer vision or 3D modeling from the user. As underlined in the previous section, 3D model of the environment is necessary to achieve high immersivity and strong augmentation of the real world. How such 3D model can be made available to the user situated in an unprepared environment? In the current mobile AR context, the classical 1D-3D (i.e. surveying/geomatics) acquisition techniques and subsequent processing display strong constraints from a cost and time standpoint. For example, topographical survey and subsequent 3D modeling can require several work days before delivering the required 3D model. User cannot wait for days on site before being able to trigger his augmentation of the world. Terrestrial LiDAR (i.e. Light Detection and Ranging) offers high potential for fast data collection and large urban coverage. However, its cost is still expensive (approximately $150,000), which limits its availability. Even if mobile LiDAR data acquisition of large scale environment increases [e.g. NAVTEQ recently adopted this technology (NAVTEQ)], a lot of urban environments have not been scanned yet and mobile LiDAR survey remains very expensive (a survey generally costs tens of thousands dollars). As a result, these techniques are not considered adapted for mobile AR applications in an unprepared environment. Other techniques has been designed especially for mobile AR applications like the working planes (Piekarski 2004) or the combination of map coordinates and inertial sensor data (Kim et al. 2007). Even if these are successful tools, they are used with specialized and expensive equipment. Therefore there is a need for 3D modeling techniques dedicated to mobile AR applications in unprepared environment. In this context it appears essential that the modeling process be low cost, fast, intuitive, ergonomic, and require neither survey equipment nor specific knowledge. Since the purpose of the 3D model is to contribute to the strong augmentation of the environment (i.e. precise positioning, occlusion management), it is important that the model displays relevant level of details and spatial precision. All the criteria aforementioned can be synthesized as the specifications the 3D modeling solution should fulfill to be relevant to mobile AR application in unprepared environment. Thus anywhere augmentation solution requires a 3D modeling solution: l l l l l l

That is affordable That provides a 3D model of item in the environment as fast as possible That operates directly on the spot That does not require preliminary training and specialized expertise That could be implemented on integrated, ergonomic, lightweight on the shelf devices That provides precise and detailed 3D models suitable for a determined augmentation level

4 Fast and Easy 3D Modeling Approach for Smartphones Relying on these specifications of the 3D modeling solution for augmented reality application in unprepared environment, we designed a fast, easy and generic approach to built 3D model using a smartphone. Smartphones provide valuable

168

V. Thomas et al.

visualization and positioning functionalities based on their integrated components. The proposed 3D modeling approach focuses currently only on buildings since these elements are major landmarks of urban environments. In addition, as manmade structures, their regular shape eases the 3D modeling process. The 3D modeling approach we designed consists of three main steps: (1) the retrieval of the 2D geographical coordinates of the buildings’ corners at the ground level; (2) the computation of the altitude at the top corners of the building; (3) the 3D modeling process and the augmentation of the reality. Each of these steps is detailed hereafter. The first step is carried out manually by the user. The 2D geographical coordinates of the buildings’ corners at ground level (Xb, Yb) are selected through a web based cartographic application. This application can rely on maps, airborne or satellite imagery. The only requirement is for the cartographic representation to display a spatial resolution relevant to the building corner precise visualization. The second step is carried out using the camera, the GPS receiver and the accelerometers. In order to compute the altitude at the top corners of the building, the user must aim with his mobile device camera at each of the previously identified building’s corners at roof level. The user can move around the building and choose the most suitable location to survey each corner. During each survey process, the GPS and accelerometers data are simultaneously recorded by the application. The elevation angle between the user and the roof corners is computed using the accelerometer data. Knowing the GPS position of the user (Xu, Yu, Zu) and the corner coordinates at ground level (Xb, Yb), the distance (d) between the user and the surveyed corner can be computed. With this distance and the elevation angle (y), the altitude difference (dz) between the user and the roof corner can be easily computed using (1). In (2), the GPS elevation data of the mobile device (Zu) is added to the altitude difference (dz) previously computed in order to assess the altitude of the corner at roof level (Zb) (cf. Fig. 2). dz ¼ d tanðyÞ

(1)

Zb ¼ Zu þ dz

(2)

The third step focuses on the 3D modeling of the building and the augmentation of the reality including the occlusion management. Since the rendering engine cannot deal with longitude and latitude angular data, all the building’s rooftop coordinates should be first transformed from a spherical geographic coordinate system to a planar coordinate system. Then, the 3D modeling of the building is carried out through a triangular meshing process relying on the 3D coordinates of the corner at ground and rooftop altitude. Since the altitudes of the corners at ground level are unknown, these values are set to 0. For now, such setting is admissible since it has no impact on the subsequent augmentation of the reality and occlusion management. In the future however, it might become a problem if texture has to be applied on the 3D model. The augmentation of the real world is performed through the superimposition of a 3D graphic element on top of the


169

Fig. 2 3D coordinates calculation of a building’s top corner

camera live video stream. The scene rendering takes into account the building 3D model location in relation to the 3D graphic element location and the user position and orientation. If the building is in the user field of view, the occlusion of the graphic element should be rendered accordingly. The 3D modeling of additional urban elements will be addressed in a near future. The diagram below synthesizes all the steps of the proposed fast and easy 3D modeling approach (Fig. 3).

5 iModelAR Prototype Relying on the approach described in the previous section, we developed a 3D modeling prototype we entitled iModelAR. iModelAR has been implemented on the iPhone 3G platform, taking full advantage of all its components (GPS, accelerometers and camera). The prototype has been programmed in the integrated development environment (IDE) XCode using Objective-C language. The three main steps of the fast and easy 3D modeling approach have been implemented as follows in iModelAR prototype: l

Step 1. The 2D geographical coordinates of the buildings’ corners at the ground level are selected using a Google maps view centered at the user’s location. In addition to placing markers at the corner location on the map, the user needs to

170

V. Thomas et al.

Zoom to current location in cartographic view

Pin down the building’s corners at ground level + the position of the 3D graphic element with markers

Render the building and the 3D graphic element corresponding to the user’s position & orientation

Transform 3D geographical coordinates to a cartographic plane projection

Legend Automatic process

Calculate the altitude of each corners at roof level

User action

Store 2D coordinates Trigger AR mode YES Select the marker number to survey

NO All corners are suveyed?

Aim with the camera view at the corresponding corner at roof level and press capture button

Store GPS + accelerometers data

Fig. 3 Citizen-based generic 3D modeling approach for smartphones

l

l

mark the location of the 3D graphic element that will augment the real world later on. (cf. Fig. 4). Step 2. The user can rely on a red cross overlaid on the camera view to help him aim precisely at the previously identified building’s corners at roof level (cf. Fig. 5). When surveying a building corner, iModelAR records series of accelerometers data. The more accelerometers data is recorded, the better the accuracy of the corner survey is but the longer it takes to complete this survey. Therefore, a trade-off is required between maximizing the total number of accelerometer records per corner and minimizing the survey time. According to our experiments, recording 250 accelerometers data seems to be the right trade-off. While accelerometer data are recorded, between five to seven GPS readings are recorded as well. These accelerometer and GPS data series are used to increase the redundancy of the orientation and position measurements. This will yield to more robust and representative position and orientation mean values. These values will be involved in the computation of the roof corner 3D coordinates as explained in the previous section. Once each corner has been surveyed, the user triggers the augmented reality view button. Step 3. The 3D coordinates of the rooftop corners are transformed from the geographical coordinates system of the GPS receiver (WGS84) to the Universal Transverse Mercator (UTM) coordinate systems relying on the WGS84 ellipsoid. The 3D triangular mesh of the building is computed in OpenGL ES


171

Fig. 4 (a) User at location, (b) Pinning down the building’s corners in the cartographic view (A); select the position of the graphic 3D element (B); user’s position (C)

Fig. 5 (a) User at location, surveying a building’s corner, (b) Surveying interface of the photographic view

172

V. Thomas et al.

Fig. 6 (a, b) Rendering of the 3D graphic element according to the user’s position and orientation while correctly managing occlusions

environment. Even though the main focus of this work was 3D modeling, an augmented reality view managing occlusions was developed to assess the relevance of the proposed approach and prototype towards anywhere augmentation. Only common lighting and color OpenGL ES functions have been used to achieve the rendering of the 3D graphic element (cf. Fig. 6).

5.1

iModelAR Performance Analysis

Within this section, the spatial precision and the fit for use of iModelAR will be presented. The impact of each of the components involved in the computation of the building’s 3D coordinates at roof level on the resulting accuracy will be assessed. The fit for use subsection will consist of the comparison between the criteria previously enounced in Sect. 3 and the capabilities of the developed application. Spatial Accuracy. Tests have been carried out according to the following testing protocol: l

l

Eight top corners on four different buildings with various heights have been surveyed using a total station. These measurements have been used as the ground truth The corners have been surveyed using iModelAR. This survey has been performed at the location of two geodesic points with known coordinates

3D Modeling for Mobile Augmented Reality in Unprepared Environment Table 1 iModelAR accuracy analysis Component Ground truth Geodesic point Xu, Yu Zu Geodesic point y Total station measurements d Total station measurements Total station measurements dz Xb, Yb Total station measurements Zb Total station measurements

l

l

Mean difference 3.0 m 3.9 m 1.3 2.3 m 1.1 m 2.0 m 4.2 m

173

Standard Deviation 2.4 m 3.1 m 0.9 1.6 m 1.0 m 2.0 m 2.9 m

Three series of observation of the eight corners have been performed at each of the two geodesic points location At the second geodesic point location, only six of the eight corners were visible

Being located on known 3D geodesic points, each position or orientation measures involved in the computation of the building corner 3D coordinates could be analyzed individually. Since the distance and the elevation difference between the geodesic points and all the corners were known precisely, the targeted measurements could be inferred. Table 1 sums up the mean difference in absolute value between the position and orientation measures recorded using iModelAR and those recorded using the total station. Horizontal Accuracy. When addressing the horizontal accuracy of a building’s corner, the main sources of error are related to the markers placement on the satellite image, the georeferencing precision and the highest zoom level available of the cartographic tiles of Google Maps. Sometimes, and because projective effects, it is tricky to approximate where the corner is at ground level if the building is tall (cf. Fig. 7) or has an extended cornice. Maximum errors up to 5 m have been assessed only for the Google Maps tile of our test zone on the Universite´ Laval campus. Regarding the georeferencing of the satellite images of Google Maps, no metadata information was available about the accuracy of this georeferencing. Therefore an uncertainty remains about the inaccuracy value that should be attached to this error source. More testing should be done at various locations to assess its impact on the horizontal precision. Another point is the maximum zoom level available for a geographic zone. A high resolution Google Maps tile will enable an iModelAR user to achieve precise placements of markers at a building’s corner. When combining the imprecision of the three aforementioned sources of errors, the mean difference between the coordinates in the (x, y) plane provided by iModelAR and those provided by the ground truth is about 2 m. Vertical Accuracy. Three components are needed to compute the altitude of a building’s roof corner with iModelAR: (1) the distance between the user and the corner that is surveyed [calculated based on their horizontal positions (x, y)]; (2) the elevation angle of the mobile device during the survey; (3) the altitude of the mobile device during the survey. Let’s focus at first on the precision of the distance between the user and the corner to survey. Two components are used to calculate this distance: the user

174

V. Thomas et al.

Fig. 7 Marker at ground level corresponding to the roof where the red cross is

position during the survey and the 2D coordinates of the marker. The horizontal position (x, y) of the user is retrieve using the iPhone embedded GPS receiver. Five to seven GPS positions are averaged each time a corner coordinate is surveyed in order to have redundancy in the data set. A mean error of about 3 m has been computed for the GPS measurement. The horizontal precision of the corner is about 2 m (cf. Horizontal Precision section). When taking into account these two accuracy values, the resulting precision for the distance was ranging between 2 and 3 m. The elevation angle is calculated based on the accelerometer data recorded during each corner survey. 250 accelerometer measures are recorded and filtered to limit the effects of the user’s quick movements. By comparing the elevation angles measured using iModelAR to the total station measurements, a mean difference of only 1.3 has been computed. When combining the elevation angle with the distance, the mean error between the building’s roof altitude as provided by iModelAR and the altitude surveyed with the total station is about 1 m. However, the iPhone’s altitude from the GPS receiver should also be involved in iModelAR computation (i.e. its altitude should be added to the building’s altitude computed by the prototype) in order to get the correct Z coordinate of the building’s roof corners. In the context of this study, the iPhone’s

3D Modeling for Mobile Augmented Reality in Unprepared Environment Table 2 List of performances of iModelAR

Criteria 3D modeling speed (4 corners building) Ease of use Cost

On the fly? With off the shelf material? Required knowledge in 3D modeling The 3D model produced can be used to augment the reality? Occlusion management?

175

iModel AR performance 0, a half space for which Ax þ By þ Cz þ D < 0, and those points which lie on the plane. As described above, computation of the point set of points lying on the plane is problematic, so this approach includes them in one or other of the half spaces. Since the method of inclusion is consistent, there is no practical or theoretic difficulty with this – as will be shown below. Using dr-rational points, a half space H(A, B, C, D) (A, B, C, D integers, M < A,B,C < M, 3M2 < D < 3M2) is defined as the set of dr-rational points p(X, Y, Z, Q), MQ X,Y,Z MQ, for which computation of the following inequalities yields these results: ðAX þ BY þ CZ þ DQÞ > 0 or ½ðAX þ BY þ CZ þ DQÞ ¼ 0 and A > 0 or

186

R.J. Thompson and P. van Oosterom

½ðBY þ CZ þ DQÞ ¼ 0 and A ¼ 0 and B > 0 or ½ðCZ þ DQÞ ¼ 0 and A ¼ 0; B ¼ 0 and C > 0: The complement of a half space is defined as: H ¼ ðA; B; C; DÞ where H ¼ ðA; B; C; DÞ: Using this definition, any plane divides space into two mutually exclusive point sets. Thus from a point-set perspective, the approach is “boundary-free”.

3.3

Convex Polytope Definition

A convex polytope is defined as the intersection of any finite number of half spaces; see examples in Figs. 5 and 6. Convex polytope C is defined as: C¼

\

Hi where fHi ; i ¼ 1::ng is a set of half spaces:

i¼1::n

In Fig. 5, the solid lines/planes are used to indicate that points which fall along the line/plane in question (AX þ BY þ CZ þ D ¼ 0), are within the convex polytope. The dashed lines/planes indicate that these points do not belong (but would belong to an adjoining polytope). Likewise, the highlighted vertices are part of the subject convex polytopes. Dotted vertices are external. Specifically, in Fig. 5, Points that lie exactly on the edge through v3 and v4 are not part of region a, but are part of b, c or e. Vertex v3, lies within region b, and v4 is within region c. The universal convex polytope, is defined as: S T C1 ¼ {} (no half spaces). Hj is defined as: C ¼ fHj g. The complement of a convex polytope C ¼ j¼1::m

v2

v1

b

v3 a

j¼1::m

c

v4 d

Convex region defined by half-planes

Convex region defined not completely bounded

Convex region defined by half-spaces in 3D

Fig. 5 Convex polytopes defined by half planes/spaces. Highlighted lines/surfaces indicate that these belong to the convex polytope (interior), dashed means outside

Integrated Representation of (Potentially Unbounded) 2D and 3D Spatial Objects

187

C4 = {H4} C1 = {H1}

C= H

C3 = {H3}

i = 1..4

C2 = {H2}

Ci

C=

i = 1..4

Convex polytope defined by half-planes

Inverse of Convex Polytope

Fig. 6 The complement of a convex polytope

Fig. 7 Definition of regular polytope from convex polytopes

C1

a

C2 b

C3

p

This is shown in Fig. 6.

3.4

Regular Polytope Definition

A regular polytope O is defined as the union of a set of convex polytopes (Fig. 7). O¼

[

Ci where Ci ; i ¼ 1::m are convex polytopes:

i¼1::m

Two special regular polytopes are defined: OF ¼ fg; O1 ¼ C1 :

188


These sets are the empty and universal sets used to define a topological space, based on definitions of the union, intersection, and complement of regular polytopes: [ i¼1;m

O

\

[

Oi ¼

Cij where Oi ¼

i¼1::m;j¼1::ni

O0 ¼ ð

[

Ci Þ \ ð

i¼1::n

O¼def

[

[

C0 j Þ ¼

ðCi

\

C0 j Þ

i¼1::n;j¼1::m

Ci where O ¼

i¼1::n

3.5

Cij

j¼1::ni

j¼1::m

\

[

[

Ci

i¼1::n

Properties of the Regular Polytope

It is relatively simple to show that the space spanned by regular polytopes is a topology (Thompson 2005b), based on the definition of regular polytope as an open set. However the space is not Euclidean. Referring to the definition of complement, it is apparent that for any regular Thus no boundary points exist between O and O polytope O,8p; p 2 O , p 2 = O. (8p; p 2 O _ p 2 O). This is in contrast with most conventional approaches where (in the mathematical model) space is partitioned into a region’s interior R , exterior R and boundary dR. A further consequence of being a boundary free representation is that the axioms of a Boolean algebra (Weisstein 1999) are satisfied (see Sect. 2.4).

4 Connectivity of Geometric Objects One of the more important properties of spatial objects is connectivity. This subject is covered in more detail in (Thompson and van Oosterom 2009). The following is a brief summary of the issues. Loosely, connectivity can be thought of as the property that an object has if it is “in one piece”. The literature contains a wide spectrum of definitions of connectivity. In the polygon representation, as defined by standards such as ISO 19107 (ISOTC211 2003), connectivity is mandated by requiring at most one outer boundary with zero or more inner boundaries, but this requires some qualification. The regions in Fig. 8 are not usually considered continuous (OGC 1999) despite having a single outer boundary. Commercial GIS, spatial DBMS products, ISO 19107 (ISO-TC211 2003) and OGC (OGC 1999) are at considerable variance in


a

b

189

c

Fig. 8 Discontinuous regions with single outer boundaries. (The circled points in A are supposed to coincide exactly, but have been drawn slightly separated for clarity)

Weak Connection

1

2

Strong Connection

3 Ca

4

5 Cb ⇒ Ca

Overlap

6 OV ⇒ Cb ⇒ Ca

Fig. 9 Modes of connectivity in 3D

their approaches to this question (van Oosterom et al. 2004; Thompson and van Oosterom 2009). It seems clear that any practical representation should support weak and strong connectivity. Following (Cohn and Varzi 1999) these are denoted: Ca if the regions touch at one or more points Cb if the regions meet at a surface (in 3D or a line in 2D) (see Fig. 9) Note that Cb ) Ca It is clear that these definitions need careful interpretation, since the definition of a line or plane of contact cannot be reliably evaluated as a point set as noted above in Sect. 2.1. It has been shown (Thompson 2005b) that definitions of Ca and Cb for regular polytopes are viable, that using either definition, the space of regular polytopes obeys the axioms of the RCC based, and that it forms a Weak Proximity Space and a Boolean Connection Algebra.

190


5 Lazy Evaluation A critical feature of this approach is that all the basic operations are computed using finite arithmetic. Thus, a half space is defined as a tuple (A, B, C, D) of integers of limited size (typically A, B, C would be 32 bit integers, D would be 64 bit). The grouping of half spaces into convex polytopes, and further into regular polytopes does not require any calculation, and so any convex polytope can be defined in purely finite integer form. In fact many of the spatial functions can be evaluated without arithmetic calculation – hence the term “Lazy Evaluation”. Dr_rational numbers are required for the calculation of overlap, continuity etc. (all the RCC operations), but it has been shown (Thompson 2005b) that the precision required for these calculations is large but fixed (if the original integers that define the half planes are as above, the calculation vertices require integers of about 128 bits. Thus it can be stated that there is a fixed limit on the maximum precision required for any calculation, and that therefore the time taken for any arithmetic operation is finite and limited. This is in contrast to the approach of using unrestricted size integer form (such as Java’s BigInteger form). This has been shown to give reasonable results in 2D calculations for the first five or so generations of calculations (Bulbul and Frank 2009), but from that time forward, the time is exponential in terms of the number of generations (increasing by a factor of about 3 per generation). This is probably acceptable in a 2D application, especially as the generations can be “reset” periodically by running an “approximate and re-validate” sweep through the database periodically (Thompson 2009). The situation is 3D is considerably worse, with each generation of calculations, the number of bits required to store coordinates increases by a factor of 10 (compared to a factor of 2 in 2D). It can be expected that the calculation times would be exponential with a similar factor (Thompson 2007).

6 Cadastral Domain This discussion is applicable to various classes of spatial information, but it has been shown to permit an elegant solution to some of the requirements of Cadastral information. Here, the fundamental item of interest is the land parcel. While these parcels are defined by measurement (survey) of their boundaries, there is no question of ownership of the boundary itself (as a mathematical line). Thus a boundary-free representation is ideal. There is a growing need to represent 3D parcels in a cadastre. These include strata parcels, units, properties overhanging roads or rail, etc. (Tarbit and Thompson 2006), but amount to a small minority of cadastral parcels in any jurisdiction. This is a strong argument for mixing 2D and 3D representations in the one database (Stoter and van Oosterom 2006), as is also required by the emerging ISO standard for the Land Administration Domain Model (LADM) (ISO-TC211 2009).


191

C A B D

Fig. 10 Mixing 2D and 3D Cadastre. Parcels A, B and D are what is usually described as “2D” parcels, but in fact are regions of space above, below and including the rectangular area. There is no explicit top or bottom to these regions – thus they are unbounded above and below. C is a volumetric parcel (fully bounded in all dimensions), with its projection onto the ground shown hashed

Fig. 11 An overview of the test region. These are primarily 2D parcels, but some 3D parcels can be glimpsed. One of these is shown in detail in the following figure

As pointed out by Stoter (2004), the so-called 2D parcels are in reality volumetric regions in space. It is merely the definition that is 2D and therefore defines a prismatic volume with undefined vertical limits. As a result it should be possible to evaluate any functions and predicates on a mixture of 2D and 3D parcels. E.g. in Fig. 10, it can be asked whether C intersects D (which it does, since it encroaches into the volume of D). Thus Cadastral data admits of a single sorted logic despite the mixture of 2D and 3D objects.

192


7 Expressiveness of the Representation In order to investigate the practicalities of the approach, a set of Java objects have been written (Thompson and van Oosterom 2006) and used to represent about a thousand 2D cadastral parcels (Fig. 11) taken from the Queensland cadastre (courtesy of the Department of Environment and Resource Management). Included in this test region were three 3D parcels. The basic functions, union, intersection and complement, and the predicates “Empty”, Ca and Cb were first implemented, using the simplifications allowed by the findings of (Thompson 2005b). The remaining functions are implemented by simple combinations of these – e.g.: ðOverlapÞ OVðp; qÞ ¼ def :Emptyðp \ qÞ (Strong external connection) ECb ðp; qÞ ¼ def Cb ðp,qÞ ^ :OV(p; qÞ The actual Java coding exactly parallels these definitions. e.g.: public boolean externallyConnectedToB(Polytope other) {return this.connectsToB(other)&& !this.overlaps(other);} Using this Java implementation, in conjunction with the mathematical proofs of consistency, it has been demonstrated that this approach, even though it does not draw on the concept of a “boundary point set”, can still supply the functionality that is usually associated with boundaries. Thus, predicates such as “in contact with”, “adjacent to”, “wholly within”, and “non-tangential proper part” are fully implemented without resorting to a definition of the type “point p is (or is not) an element of the boundary of” the regions. On the other side, it is still possible to generate a set of flat polygons that approximate to the boundary of the regular polytopes for presentation purposes. The remainder of this section shows the kind of data that has been represented, and its presentation (Sect. 7.1) and discusses the storage requirements of the approach (Sect. 7.2).

7.1

Presentation

2D cadastral parcels are unbounded above and below (as described in Sect. 6), which makes them impossible to display as a set of bounding flat polygons. Therefore in Figs. 11 and 12, 2D regular polytopes have been displayed with “fences” to indicate the position of the vertical boundaries. The 3D parcels are displayed as a set of flat polygons defining their outer surfaces. The Java coding has type-casting that allow any of the 2D objects to be cast to their 3D equivalent – thus allowing any of the functions to be applied to any combination of features. For example. H(A, B, D) a 2D half plane is cast to H(A, B, 0, D), a 3D half space oriented vertically. The fact that the 2D object can be represented in 3D without


193

Fig. 12 3D parcel amongst 2D parcels (Parcel A and 2D parcel E comprise a restaurant. Parcel overhangs the roadway represented by parcels B, C and D)

Table 1 Storage requirements

Conventional Vertex representation Regular polytope representation

2D 3D Simple Moderate Complex Simple Moderate Complex 96 b 864 b 80 kb 404 b 5,668 b 560 kb

Without topology With topology 148 b Basic 144 b approach Shared half 147 b spaces

532 b 1,860 b

40 kb 165 kb

230 b 212 b

1,736 b 2,344 b

160 kb 207 kb

1,927 b

172 kb

196 b

2,148 b

187 kb

resorting to an artificial “top” and “bottom” is an attractive feature, which also fits to reality, as legally the parcels have no defined lower or upper bound. Figure 12 illustrates a case where mixed dimensionality predicates and functions can be of value, given a set of basic functions that are rigorous and therefore can be combined with confidence. E.g. “find the parcels that are encroached by parcel A” (“OV(A)”); “find the parcels that are strongly connected to but do not overlap E” (“ECb(E)”); and “is A entirely within B, C union D, but in strong contact with the edge” (“A.TPPb(B \ C \ D)”). These types of function can be used to ensure correctness of data, and in a practical database implementation could be specified as constraints (such as that parcels of a particular type cannot overlap one another).

194

7.2


Storage Requirements

Based on the Java classes, and on theoretical considerations, the storage requirements of the regular polytope approach has been compared with the more conventional (vertex-based) representations of simple, moderate or complex objects. In general the requirements are quite comparable in 3D (but in 2D, the conventional approach has significantly lighter requirements). Table 1 shows a summary of the comparison. More detail may be found in Thompson (2007). The shared half space approach allows that any half space that is part of the definition of more than one convex polytope is stored once only. Thus a form of adjacency topology is stored in the database, saving storage, and permitting faster calculation of connectivity.

8 Conclusions The major contribution of this approach is that it supports a rigorous algebra of topology and connectivity in 2D and 3D spatial applications. This has been shown by axiomatic proof (Thompson 2005b), has been verified to be implementable as a set of objects and supporting algorithms, and tested on a set of real features of useful size. In addition to being applicable to 2D and 3D spaces, it has also been shown to be particularly useful in the cadastral applications by allowing the mixing of representations of different dimensionality without jeopardising the rigour. Other important applications will benefit from this approach. For example, in topography a 2.5D land surface could be represented by a set of convex polytopes with vertical sides, an upper bound (coinciding with the terrain surface) and no lower bounds wherever a TIN or DEM would be applicable, but with fully general (3D) convex polytope decomposition where a full 3D TEN or other construct would be needed (Verbree et al. 2005). For example, in Fig. 13, the majority of the land

Convex Polytopes, unbounded below

Fig. 13 Region of 2.5D surface meeting 3D structure

Bounded Convex Polytopes


195

surface can be represented as convex polytopes bounded above only (unbounded below), while the bridge is composed of a collection of fully bounded convex polytopes. Clearly there are advantages gained by allowing rigorous and correct operations between these two types of data (Cadastral and Topographic). The distinguishing feature of this approach is that it de-emphasises the boundary of regions, and eliminates the concept of a set of boundary points. In doing do, it does not lose the functionality that is normally associated with the boundary representations, but provides that functionality in rigorously computable form.

8.1

Further Research

At present, there is no provision for the support of representations that use lowerdimensional geometries, such as roads that are so narrow at the scale of representation that they are stored as linear geometry. Extension into lower dimension features (surfaces, lines and points) is not problematic, but does highlight the difficulty of mixing paradigms – in particular any attempt to treat lines or surfaces as point sets. Thus a sorted algebra such as ROSE (extended to 3D) will be needed. Note that at the current level of development, this approach has not been highly optimized. In fact, the performance of spatial queries and updates does not compete with the highly efficient techniques on the current market. This is an area that would benefit from further research.

References Angel, E. (2006). Interactive Computer Graphics: A Top-Down Approach Using OpenGL. Boston, MA, Addison Wesley. Bulbul, R. and A. U. Frank (2009). BigIntegers for GIS: Testing the Viability of Arbitrary Precision Arithmetic for GIS Geometry. 12th AGILE Conference, Hannover, Germany. Cohn, A. G. and A. C. Varzi (1999). Modes of Connection. Spatial Information Theory. Proceedings of the Fourth International Conference. Berlin and Heidelberg, Springer-Verlag. Courant, R. and H. Robbins (1996). The Denumerability of the Rational Number and the NonDenumerability of the Continuum. What Is Mathematics?: An Elementary Approach to Ideas and Methods. Oxford, Oxford University Press: 79–83. Coxeter, H. S. M. (1974). Projective Geometry. New York, Springer-Verlag. D€ untsch, I. and M. Winter (2004). Algebraization and representation of mereotopological structures. Relational Methods in Computer Science 1: 161–180. Egenhofer, M. J. (1994). Deriving the composition of binary topological relations. Journal of Visual Languages and Computing 5(2): 133–149. Franklin, W. R. (1984). Cartographic errors symptomatic of underlying algebra problems. International Symposium on Spatial Data Handling, Zurich, Switzerland: 190–208. Gaal, S. A. (1964). Point Set Topology. New York, Academic Press. G€uting, R. H. and M. Schneider (1993). Realms: a foundation for spatial data types in database systems. 3rd International Symposium on Large Spatial Databases (SSD), Singapore. ISO-TC211 (2003, 2001-11-21). Geographic Information – Spatial Schema. ISO19107, from http://www.iso.org/iso/catalogue_detail.htm?csnumber¼26012

196


ISO-TC211 (2009). Geographic Information – Land Administration Domain Model (LADM). ISO/CD 19152. Lema, J. A. C. and R. H. G€ uting (2002). Dual grid: a new approach for robust spatial algebra implementation. GeoInformatica 6(1): 57–76. Lemon, O. and I. Pratt (1998). Complete logics for QSR [qualitative spatial reasoning]: a guide to plane mereotopology. Journal of Visual Languages and Computing 9: 5–21. Mehlhorn, K. and S. N€aher (1999). LEDA: A Platform for Combinatorial and Geometric Computing. Cambridge, UK, Cambridge University Press. Naimpally, S. A. and B. D. Warrack (1970). Proximity Spaces. Cambridge, Cambridge University Press. Nunes, J. (1991). Geographic Space as a Set of Concrete Geographic Entities. Cognitive and Linguistic Aspects of Geographic Space. D. M. Mark and A. Frank (eds.). Dordrecht, Kluwer Academic: 9–33. OGC (1999, 5 May 1999). Open GIS Simple features Specification for SQL. Revision 1.1. Retrieved 15 Oct 2003, from http://www.opengis.org/specs/ Randell, D. A., et al. (1992). A Spatial Logic Based on Regions and Connection. 3rd International Conference on Principles of Knowledge Representation and Reasoning. Cambridge, MA, Morgan Kaufmann. Roy, A. J. and J. G. Stell (2002). A Qualitative Account of Discrete Space. GIScience 2002, Boulder, CO, USA. Stoter, J. (2004). 3D Cadastre. Delft, Delft University of Technology. Stoter, J. and P. van Oosterom (2006). 3D Cadastre in an International Context. Boca Raton, FL, Taylor & Francis. Tarbit, S. and R. J. Thompson (2006). Future Trends for Modern DCDB’s, a new Vision for an Existing Infrastructure. Combined 5th Trans Tasman Survey Conference and 2nd Queensland Spatial Industry Conference, Cairns, Queensland, Australia. Thompson, R. J. (2005a). 3D Framework for Robust Digital Spatial Models. Large-Scale 3D Data Integration. S. Zlatanova and D. Prosperi. Boca Raton, FL, Taylor & Francis. Thompson, R. J. (2005b). Proofs of Assertions in the Investigation of the Regular Polytope. Retrieved 2 Feb 2007, from http://www.gdmc.nl/publications/reports/GISt41.pdf Thompson, R. J. (2007). Towards a Rigorous Logic for Spatial Data Representation. PhD thesis, Delft University of Technology, Delft, The Netherlands, Netherlands Geodetic Commission. Thompson, R. J. (2009). Use of Finite Arithmetic in 3D Spatial Databases. 3D GeoInfo 08, Seoul, Springer. Thompson, R. J. and P. van Oosterom (2006). Implementation Issues in the Storage of Spatial Data As Regular Polytopes. UDMS 06, Aalborg. Thompson, R. J. and P. van Oosterom (2009). Connectivity in the Regular Polytope Representation. GeoInformatica, October: 24. van Oosterom, P., et al. (2004). About Invalid, Valid and Clean Polygons. Developments in Spatial Data Handling. P. F. Fisher (ed.). New York, Springer-Verlag: 1–16. Verbree, E., et al. (2005). Overlay of 3D features within a tetrahedral mesh: A complex algorithm made simple. Auto Carto 2005, Las Vegas, NV. Weisstein, E. W. (1999). Boolean Algebra. MathWorld – A Wolfram Web Resource. Retrieved 20 Jan 2007, from http://mathworld.wolfram.com/BooleanAlgebra.html Zlatanova, S. (2000). 3D GIS for Urban Development. Graz, Graz University of Technology. Zlatanova, S., et al. (2002). Topology for 3D spatial objects. International Symposium and Exhibition on Geoinformation, Kuala Lumpur.

Interactive Rendering Techniques for Highlighting in 3D Geovirtual Environments Matthias Trapp, Christian Beesk, Sebastian Pasewaldt, and J€ urgen D€ollner

Abstract 3D geovirtual environments (GeoVE), such as virtual 3D city and landscape models become an important tool for the visualization of geospatial information. Highlighting is an important component within a visualization framework and is essential for the user interaction within many applications. It enables the user to easily perceive active or selected objects in the context of the current interaction task. With respect to 3D GeoVE, it has a number of applications, such as the visualization of user selections, data base queries, as well as navigation aid by highlighting way points, routes, or to guide the user attention. The geometrical complexity of 3D GeoVE often requires specialized rendering techniques for the real-time image synthesis. This paper presents a framework that unifies various highlighting techniques and is especially suitable for the interactive rendering 3D GeoVE of high geometrical complexity.

1 Introduction Highlighting of objects is an important visualization principle used in human computer interaction. This form of visual feedback can be considered as an instance of higher-order visualization (Bj€ ork et al. 1999), similar to the principle of focus þ context visualization (Cockburn et al. 2008). Usually, a state of an object different from the original one is communicated by displaying it in a different visual style or appearance. These distinguishable styles, such as color overlay or outline, yields support of the pre-attentive cognition. Despite changes of color or texture, also geometrical properties such as scale can be used to highlight important objects (Glander and D€ollner 2009). A state transition can be initiated manually, e.g., by selecting one or multiple objects, or by the result of a computation, e.g., a database query. To summarize, object highlighting has a number of different applications:

M. Trapp (*), C. Beesk, S. Pasewaldt, and J. D€ ollner Hasso-Plattner-Institute, University of Potsdam, Prof-Dr.-Helmert-Str. 2-3, 14482 Potsdam, Germany e-mail: [email protected]


197

198

M. Trapp et al.

User Selection Preview. This is the classical use case for object highlighting. A number of selected objects are rendered using different styles. Usually the user modifies only a single object, which is in the users focus. Visualization of Computational Results. In contrast to a manual selection, a number of objects can be the result of a computation, such as a database query. Instead of showing the results in a list, applied object highlight has the advantage, that the human can spatially cluster the highlighted result immediately. Navigation Aid. Highlighting techniques can also be applied to landmarks, points-of-interest (POI), routes, as well as to navigation way points in order to guide the users attention. In Omer et al. (2006) the highlighting of scene elements, such as local and global landmarks, is considered as navigation aid that “clearly helps improve orientation in the virtual model of a real city”. In Jobst and D€ollner (2006) it is argued that, despite building aggregation and simplification, appropriate highlighting can compensate the dead value areas in virtual 3D city models. Further, Bogdahn and Coors state that “highlighting of complete buildings using false colors might be a first step. However, in a dense urban environment and for pedestrians it could be necessary to provide more sophisticated visual hints, like highlighting the correct entrance to a big building or a building complex” (Bogdahn and Coors 2010). Robinson documented the application of highlighting techniques for information visualization in 2D GeoVE (Robinson 2006). With respect to 3D GeoVE, the applications and technical implications of highlighting techniques is widely not researched. Besides the usually high geometrical complexity of 3D GeoVE, a highlighting technique has to approach the following characteristics of virtual environments: Object Size and Shape. In virtual city models, object size and shape can differ enormously. The size can vary from big (buildings) to very small (e.g., pedestrians, city furniture, etc.). Some shapes can have extreme spatial extent along only one major axis, thus, are not fully visible on the viewport. We can further distinguish between convex and non-convex shapes. Number of Objects. Depending on the application, the visualization technique has to highlight a number of different objects simultaneously. In a simple case, only one object has to be highlighted, but in a general use case a large number of objects are encountered. For instance, this case often occurs when editing virtual city models. Camera Orientation and Perspective. If unconstrained navigation and interaction metaphors are used in virtual environments, a system has to handle different perspectives and orientations of the virtual camera, which have an influence on object occlusion and the objects size on the viewport. This paper presents applications of existing highlighting rendering techniques to 3D GeoVE. It provides a form of categorization and introduces an extensible rendering pipeline that enables an unified real-time rendering process, which can be easily integrated into existing rendering systems and applications. This paper is structured as follows. Section 2 discusses origins of different highlighting techniques and related work concerning the topic of highlighting. Section 3 presents a conceptual overview of our framework and briefly discuss details of our

Interactive Rendering Techniques for Highlighting in 3D Geovirtual Environments

199

prototypical implementation. Section 4 shows application examples, presents a comparison between highlighting techniques, as well as ideas for future work. Section 5 concludes this work.

2 Review of Object-Highlighting Techniques In this paper, we focus on highlighting of objects that are located on screen. In contrast to approaches for off-screen location visualization (Trapp et al. 2009), onscreen highlighting techniques should enable the estimation of the 3D position, as well as the dimensions of an object. There are a number of different approaches for 3D object highlighting and selection preview, which are mostly inferred from 2D visualization. In early stages of 2D computer graphics, different visual representations of objects are used, i.e., sprites or billboards (Akenine-M€ oller et al. 2008) with different hues. Such an approach is not appropriate for current large-scale data set common in GeoVE. This section propose a classification of existing highlighting approaches that mainly distinguishes between three types of rendering techniques: Style-variance techniques are based on modifying the appearance of an object that results in an obvious distinction from the rest of the scene. The highlighting hint depends directly on the object and is therefore referred to as direct hint. Outlining techniques achieve this effect by enhancing the outline or silhouette of an object. We denote this setting as attached hint. Glyph-based techniques rely on icons or glyphs that are attached to the object to be highlighted. These kind of techniques deliver an indirect hint to the user. Figure 1 provides examples of these techniques applied to a single object within a virtual 3D city model. The remainder of this section discusses these types in detail. Related and in addition to the categories above, highlighting can also be achieved using cartographic generalization operations as described in Sester (2002).

2.1

Style-Variance Techniques

Probably the most widespread highlighting techniques are instances of style variance techniques. The highlighting effect is achieved by modifying the appearance in which an object or scene is usually depicted. Such appearance modification can be arbitrarily complex, or as simple as overdrawing the same object geometry with a dominant color that can easily be distinguished from all other colors in the depicted scene. Modifications can be applied to an object or area that should be highlighted (focus-based) or to the remaining scene (context-based). Focus-Based Style Variance Techniques. In 3D virtual environments, focusbased style variance techniques are suitable to be applied to objects that are not occluded. Besides color overlay (Fig. 1c), the appearance can be modified by using

200

a.1

c

M. Trapp et al.

a.2

b.1

d

b.2

e

Fig. 1 Overview of different highlighting techniques applied to a single building within a virtual 3D city model: (a): outlining techniques (1: simple outline, 2: glow as generalized version of outlines) (b): context-based style variances (1: vignetting, 2: semantic depth-of-field) (c): focusbased style variances (color overlay) (d and e: glyph-based object highlighting

different rendering styles, such as wire frame rendering known from standard 3D modeling tools. Another classic rendering approach, which was often used in 2D sprite-based games, enfolds different representations of the same sprite/billboard (Akenine-M€oller et al. 2008). Due to the potentially massive amount of additional data, the usage of different geometric or image-based representations of each single object is hardly manageable in large-scale 3D GeoVE. Therefore, the style invariance is created dynamically, e.g., by blending the standard appearance with a highlighting color. More advanced rendering techniques can be used, such as real-time non-photorealistic rendering (NPR) effects (Cole et al. 2006) that can be controlled manually by an artist or automatically via style transfer functions (Bruckner and Gr€oller 2007). Context-Based Style Invariance. Techniques of this category convey the appearance of the objects to highlight, while modifying the surrounding scene context in a way that the objects-of-interests are emphasized. Figure 2 shows examples of this category of highlighting techniques: vignetting and semantic depth-of-field (SDOF) (Kosara et al. 2001, 2002). In Cole et al. (2006), an advanced technique is described that modifies the quality and intensity of edges, color saturation, and contrast.


a

b

c

d

201

Fig. 2 Vignetting using gray-scale (a) and contrast and gamma modifications (b), as well as semantic depth-of-field (c and d), as examples of context-based style variance techniques for highlighting

2.2

Outlining Techniques

Depending on the application, the change of style is not always appropriate or possible. Especially in 3D GeoVE, it can be desirable to convey the appearance of the object and the surrounding scene, such as its facade information which could be essential for orientation and navigation. In such use-cases, outlining techniques can be applied that enhance or modify the contours or silhouettes of an object only. Such enhanced contour can possess different line styles and sizes. For real-time rendering purposes, these silhouettes can be rendered using image-based (Nienhaus and D€ollner 2003) or the more recent geometry-based (Hermosilla and Vzquez 2009) approaches. Figure 1c shows the application of image based-glow (O’Rorke and James 2004) as a general concept for outlining objects. The attached hint is one of the major advantage of outline techniques: an increased visibility can be gained by increasing the thicknesses of the outline. But increased thickness of the hint also introduces occlusion of the surrounding object area. This can be compensated partially by using a drop-off function, which results in smaller occlusion then occurred for the alternative glyph-based techniques. The application of glow can be considered as a generalized variant of the outlining technique.

202

2.3

M. Trapp et al.

Glyph-Based Techniques

Another technique performs highlighting by using additional glyphs or icons, which are placed on top or aside an object, depending on the perspective. The highlighting effect is achieved by the difference between presence or absence of a glyph. This technique is frequently used in strategic games or games employing a third person perspective. Here, usually the orientation of the virtual camera is fixed or limited. We can distinguish between static and dynamic glyphs, e.g., visugrams introduced in Fuchs et al. (2004). The latter one includes additional information about the status of an object. A simple example using the objects axis-aligned bounding box (AABB) as a glyph is depicted in Fig. 1e.

3 Interactive Rendering of Highlighting Techniques This section introduces the basics of a highlighting framework for 3D GeoVE as well as a fully hardware accelerated, prototypical implementation for raster-based graphics. We use OpenGL and OpenGL shading language (GLSL) (Kessenich 2009) for the presented implementation that is mainly based on image-based representations of the 3D scene (Saito and Takahashi 1990; Eissele et al. 2004). This has the advantage that the techniques described in Sect. 2 can be applied on an uniform basis. Figure 3 shows an overview of our implementation pipeline for the real-time image synthesis. Basically, it consists of the following three components: 1. Image Generation. This step forms the basis for the image-based separation between objects or areas to highlight (focus) and the remaining scene (context). Figure 4 shows the data that is required for the two subsequent steps (Sect. 3.1). 2. Mask Processing. To enable smooth transitions between focus and context regions, this step applies image-based post-processing methods, such as jumpflooding and convolution filtering to the mask texture (Sect. 3.2).

Highlighting Parameterization

Scene Textures (Fig. 4)

3D Scene Geometry

Mask Texture

Processed Mask

Final Image

...

2D/3D Proxy Geometry

Image Generation

Mask Processing

Highlighting Technique

Fig. 3 Conceptual overview of the rendering pipeline for applying highlighting techniques to 3D GeoVE in real-time The data flow is depicted using stippled red and the control flow uses solid violet lines


a

b

c

203

d

Fig. 4 Image-based representations (scene textures) of the 3D scene that are required by the highlighting pipeline: color (a), depth (b), color coded object IDs (c) and mask (d) values are derived within a single rendering pass

3. Application of Highlighting Techniques. In this final step, style invariance, outline, and glyph-based highlighting techniques are applied to the scene texture. The result is then written into the frame buffer (Sect. 3.3). As input, the pipeline takes textured polygonal geometry. It requires an unique identifier per scene object. To increase rendering performance, we represents each numerical object ID as per-vertex attribute of the specific object mesh. Under the assumption that static geometry is used, this procedure enables geometry batching (Akenine-M€oller et al. 2008) or streaming without modifying the proposed rendering pipeline. Therefore, the presented approach is suitable especially for real-time rendering of geometrical complex 3D scenes, since the geometry is rasterized only once.

3.1

Image Generation

Image generation represents the first step in the highlighting pipeline and enables the separation of focus and context regions. During a single rendering pass, imagebased representations (at viewport resolution) of the 3D geometry (scene textures) and a mask texture is created (Fig. 4). Therefore, render-to-texture (RTT) in combination with fragment shaders and multiple render targets (MRT) is used, which enables to write fragments into multiple raster buffers simultaneously. The generated mask texture (Fig. 4d) contains focus regions (white) and context regions (black). Our framework distinguishes between two methods for representing the input for the mask generation step: Scene Objects. If complete scene objects (e.g., buildings) are selected for highlighting, their fragments are taken as input for generating the mask texture (Fig. 4d). Their respective object IDs are encoded in an ID buffer (Fig. 4c). This approach works only for highlighting complete objects (meshes). If it is required to highlight only parts of a single object or multiple objects, as well as highlighting regions of 3D scene that have no geometrical representation, proxy objects have to be used. Proxy Objects. These objects are presented by additional polygonal geometry that is located in the 3D scene or on the viewport. They can operate as proxy for

204

M. Trapp et al.

objects which screen size would be too small (e.g., street furniture), thus, a highlighting effect would hardly be perceivable by a user. Such a proxy object can be generated automatically, i.e., if the screen area of the projected bounding representation (e.g., AABB) an object is below a threshold, proxy geometry with a sophisticated on-screen size is created. Another use case for proxies is the partially highlighting of objects or the highlighting of routes affecting multiple objects. Here, the proxy geometry is usually created manually by the user (Fig. 5a–c), e.g., by directly painting the proxy shapes on the viewport. The resulting line segments are converted to a polygonal representation, which is then used as input for the mask generation step. Later on, the user can modify the position, scale, rotation of a proxy interactively. Note that the color and depth value of proxy objects are not written into the color and respective depth texture. Despite approximations of small objects, proxy shapes can be used for implementing Magic Lenses (Bier et al. 1993). Our system mainly distinguished between 2D and 3D lenses. 2D lenses are located in screen space and move with the virtual camera. They can be used to implement auto-focus features as described in Hillaire et al. (2008a, b) (Fig. 5f). The proxy geometry of 3D lenses is placed in the 3D scene and does not align with the virtual camera (Fig. 5d, e).

3.2

Mask Processing

After the mask texture is created, the mask processing step is performed. It basically enables the creation of smooth transition between the focus and context areas within the mask texture. The mask processing is implemented using RTT in

a

b

c

d

f

e

Fig. 5 Application of proxy-geometry for defining focus regions [cylindrical (a), free-form (b), multiple proxies (c)]. (d): Application examples of semantic depth-of-field using a 3D magic lens (d and e) and a 2D viewport lens (f) displayed with a white frame


205

combination with multi-pass rendering (G€ oddeke 2005). We basically apply two different processing algorithms: jump-flooding and convolution filtering. The jump-flooding algorithm (Rong and Tan 2006) is used to perform distance transforms between focus and context areas. If only a small dilation of the input mask is required, e.g., for creating the glow outline, we apply convolution filtering instead. The final value of the mask can be controlled by a global drop-off function. It weakens or exaggerates the previous results from jump flooding or convolution filtering.

3.3

Application of Highlighting Techniques

After the processing of the mask texture is finished, the final pipeline phase applies the respective focus- or context-base style variance and outline highlighting techniques at a per-fragment level, using fragment shader. Therefore, the color values in the scene texture (Fig. 4a) are used to apply color overlay or vignetting based on the values of the mask texture. Given the object ID texture (Fig. 4c), each object can be highlighted using a different technique. In general, the intensity of an effect, e.g., blur or opaqueness of a highlighting color, is controlled by the values of the mask texture. To implement SDOF we apply convolution filtering with different kernels in combination with multi-pass rendering. The trade-off between rendering speed and output quality can be controlled by the choice of the filter kernel. Gaussian blur requires more rendering passes than a box filter, but delivers a better blur quality. Subsequently, the resulting texture of this previous step is applied to a screenaligned quad that is then rendered into the frame buffer. The stored depth values (Fig. 4b) are also written into the frame buffer. Finally, the glyph-based highlighting techniques, such as bounding boxes or arrows are applied to the frame buffer, by using standard forward rendering.

4 Results and Discussion This section discusses the presented highlighting techniques and their implementation using different application examples, by providing a comparison, as well as ideas for possible future work. We tested our rendering techniques using data sets of different geometrical complexity: the generalized model of Berlin comprised 1,036,322 faces, the model of the Grand Canyon 1,048,560 faces, and the artificial 3D city model contains 34,596 faces. The performance tests are conducted using a NVIDIA GeForce GTX 285 GPU with 2048 MB video RAM on a Intel Xeon CPU with 2.33 GHz and 3 GB of main memory. The 3D scene geometry was not batched and no view frustum and

206

M. Trapp et al.

occlusion culling are applied. We are able to render all depicted scenes in real-time, i.e., within the range of 12–32 frames-per-second. The performance of the presented image-based approach is geometry-bounds by the geometrical complexity of the 3D scene and fill-limited by the number of mask-processing passes, which have to be performed at the viewport resolution.

4.1

Application Examples

Highlighting as a basic technology has various fields of application. Figure 6 shows possible application examples for the presented rendering techniques within 3D GeoVE. Besides the highlighting of single objects, we focus on route highlighting and the visualization of computational results. Route and Landmark Highlighting. Figure 6a shows a combination of contextand focus-based style variance techniques applied to a generalized version of a virtual 3D city model of Berlin (Glander and D€ollner 2009). Here, a vignetting technique is used to highlight the route and color-highlighting is applied to the start and end position of the route. Additional, important landmarks that are nearby the route are highlighted in orange. Visualization of Computational Results. Certain highlighting variants, such as color or glyph-based techniques, can be used for information visualization within 3D GeoVE. Figure 6b shows a possible visualization of a data base query result. A color-overlay technique is used to categorize a number of buildings. Here, the color blended over the facade textures can be used to encode specific data values. Object Highlighting. Figure 6c shows the application of the focus-based style variance operator to the visualization of digital cultural heritage artifacts. The findings of a basement are highlighted in yellow to delimit it from the remaining artifacts, which base color is orange. Additional to coloring, we applied Gooch shading (Gooch et al. 1998) and unsharp masking the depth buffer (Luft et al. 2006) to improve the perceptibility of the scene objects.

a

b

c

Fig. 6 Application examples of different highlighting techniques applied to 3D GeoVE. (a): Highlighting as navigation aid for routing purposes within a generalized virtual 3D city model of Berlin. (b): Using coloring for building categorization in an artificial virtual city model. (c): Highlighting of finding groups within an interactive digital cultural heritage application


4.2

207

Comparison of Highlighting Techniques

Figure 7 shows a comparison of style-variance, outline, and glyph-based highlighting techniques with respect to object occlusion and depth-cues. Although color- and glyph-based highlighting techniques are well established techniques to put focus on objects, they suffer from disadvantages: in the case of color highlighting, the appearance of the building is altered because the facade is dyed. As a result, important information, such as building color or texture details, become hardly recognizable. Further, in the field of city model analysis, the color attribute is often used to express a membership to semantic groups. Hence, color highlighting becomes unfeasible. Another critical problem in 3D virtual environments is occlusion. For example, if the POI is covered by another building, the viewer does not get any visual hints to guide his or her attention (Fig. 7a). In contrast to that, glyphs neither change texture information nor the buildings appearance. Instead, an additional geometric feature (glyph) is introduced atop of the object, e.g., in the form of a billboard. The size and the position of glyphs can be adapted dynamically, to avoid occlusion and ensure visibility (Fig. 7c). One disadvantage of this method is missing depth cues, due to missing scale hints. If the POI is occluded by additional geometry and the scene is displayed in a pedestrian view, the user can hardly distinguish to which object the glyph belongs to (Fig. 7e). Using an context-based or outline highlighting technique as a method of object emphasis seems to be a promising approach. First, no relevant appearance parameters of the objects are changed. Second, even if objects are partly or completely occluded, an outline or glow can still be recognized to a certain degree (Fig. 7b). Problems of mapping, as described above, are reduced because the outline can be seen as a propagated building silhouette and is, therefore, view invariant as well as supports an acceptable depth cue (Fig. 7d). One drawback of huge opaque outlines is that they can reach into other objects, which may lead to an unwanted change of facade information. The application of SDOF to 3D GeoVE exhibits a number of problems. If applied to virtual 3D city model the user is often distracted while trying to focus the blurred areas. In the case of 3D landscape models, viewed from a birds-eye perspective, this effect appears less stronger. However, semantic depth-of-field fails if the screen size of a highlighted object is to small.

a

b

c

d

e

Fig. 7 Comparison of style-variance, outline, and glyph-based highlighting techniques with respect to occlusion (a–c) and the objects depth cue (d and e)

208

4.3

M. Trapp et al.

Challenges for Future Work

There are a number of possibilities for extending the presented work. We strive to extend the design space of highlighting by using NPR techniques for the stylization of 3D GeoVE (D€ollner et al. 2005). Of particular interest is the question, if the visual differences between photo-realistic and NPR techniques are sufficient for the application as highlighting technique. We further like to investigate the computation of specific highlighting color sets, i.e. given the image-based representation of a virtual scene, what set of colors have the maximal visual differences, and how can they be computed automatically? Possible future work can furthermore comprise a feasibility study that evaluates if the presented rendering technique could be used for visualization on mobile devices.

5 Conclusions This paper presents a framework for applying highlighting techniques to multiple objects in 3D GeoVE of high geometrical complexity. We further present rendering techniques to perform real-time. Further, a GPU based implementation has been presented that enables the image synthesis for various highlighting techniques at interactive frame rates. Furthermore, different highlighting-techniques have been compared with respect to 3D GeoVE. It has been shown, that the capabilities of outline- and context-based style variant techniques support the preemptive perception, while dealing with disadvantages of other highlighting techniques. Our framework can easily be integrated into existing software products, so it could be a promising addition to existing focus+context visualizations. Acknowledgments This work has been funded by the German Federal Ministry of Education and Research (BMBF) as part of the InnoProfile research group “3D Geoinformation” (www.3dgi.de). We like to thank Tassilo Glander for providing the generalized virtual 3D city model of Berlin.

References Akenine-M€oller, T., Haines, E., Hoffman, N.: Real-Time Rendering, 3rd Edition. A. K. Peters, Ltd., Natick (2008) Bier, E.A., Stone, M.C., Pier, K., Buxton, W., Derose, T.D.: Toolglass and magic lenses: The seethrough interface. In: SIGGRAPH ’93: Proceedings of the 20th annual Conference on Computer Graphics and Interactive Techniques, New York, ACM Press (1993) 73–80 Bj€ork, S., Holmquist, L.E., Redstr€ om, J.: A framework for focus+context visualization. In: Proceedings of INFOVIS ’99, Salt Lake City (1999) 53 Bogdahn, J., Coors, V.: Using 3d urban models for pedestrian navigation support. In: GeoWeb 2010 (July 2010) Bruckner, S., Gr€oller, M.E.: Style transfer functions for illustrative volume rendering. Computer Graphics Forum 26(3) (September 2007) 715–724 was awarded the 3rd Best Paper Award at Eurographics 2007


209

Cockburn, A., Karlson, A., Bederson, B.B.: A review of overview+detail, zooming, and focus +context interfaces. ACM Computing Surveys 41(1) (2008) 1–31 Cole, F., DeCarlo, D., Finkelstein, A., Kin, K., Morley, K., Santella, A.: Directing gaze in 3D models with stylized focus. In: Eurographics Symposium on Rendering. Vol. 17 (2006) 377–387 D€ollner, J., Buchholz, H., Nienhaus, M., Kirsch, F.: Illustrative visualization of 3d city models. In Erbacher, R.F., Roberts, Jonathan C. and Gr€ ohn, M.T., B€orner, K., eds.: Visualization and Data Analysis. Proceedings of the SPIE, San Jose, International Society for Optical Engine (SPIE). Vol. 5669 (2005) 42–51 Eissele, M., Weiskopf, D., Ertl, T.: The g2-buffer framework. In: Tagungsband SimVis ’04, Magdeburg (2004) 287–298 Fuchs, G., Kreuseler, M., Schumann, H.: Extended focus and context for visualizing abstract data on maps. In: CODATA Prague Workshop on Information Visualization, Presentation, and Design, Czech Technical University in Prague, Prague, The Czech Republic (March 2004) Glander, T., D€ollner, J.: Abstract representations for interactive visualization of virtual 3d city models. Computers, Environment and Urban Systems 33(5) (2009) 375–387 G€oddeke, D.: Playing ping pong with render-to-texture. Technical report, University of Dortmund, Germany (2005) Gooch, A., Gooch, B., Shirley, P., Cohen, E.: A non-photorealistic lighting model for automatic technical illustration. In: SIGGRAPH, New York (1998) 447–452 Hermosilla, P., Vzquez, P.: Single pass GPU stylized edges. In Sern, F., Rodrguez, O., Rodrguez, J., Coto, E., eds.: IV Iberoamerican Symposium in Computer Graphics (SIACG), Margarita Island, Venezuela, June 15–17 (2009) Hillaire, S., Lcuyer, A., Cozot, R., Casiez, G.: Depth-of-field blur effects for first-person navigation in virtual environments. IEEE Computer Graphics and Applications 28(6) (2008) 47–55 Hillaire, S., Lećuyer, A., Cozot, R., Casiez, G.: Using an eye-tracking system to improve camera motions and depth-of-field blur effects in virtual environments. In: IEEE International Conference on Virtual Reality (IEEE VR), Reno (2008) 47–50 Jobst, M., D€ollner, J.: 3d city model visualization with cartography-oriented design. In Schrenk, M., Popovich, V.V., Dirk Engelke, P.E., eds.: 13th International Conference on Urban Planning, Regional Development and Information Society. REAL CORP, Vienna (May 2008) 507–515 Kessenich, J.: The OpenGL Shading Language. The Khronos Group Inc., Beaverton (2009) Kosara, R., Miksch, S., Hauser, H.: Semantic depth of field. In: INFOVIS ’01: Proceedings of the IEEE Symposium on Information Visualization 2001 (INFOVIS’01), Washington, DC, IEEE Computer Society (2001) 97 Kosara, R., Miksch, S., Hauser, H., Schrammel, J., Giller, V., Tscheligi, M.: Useful properties of semantic depth of field for better f+c visualization. In: VISSYM ’02: Proceedings of the Symposium on Data Visualisation 2002, Aire-la-Ville, Switzerland, Eurographics Association (2002) 205–210 Luft, T., Colditz, C., Deussen, O.: Image enhancement by unsharp masking the depth buffer. ACM Transactions on Graphics 25(3) (July 2006) 1206–1213 Nienhaus, M., D€ollner, J.: Edge-enhancement – an algorithm for real-time non-photorealistic rendering. International Winter School of Computer Graphics, Journal of WSCG 11(2) (2003) 346–353 Omer, I., Goldblatt, R., Talmor, K., Roz, A.: Enhancing the Legibility of Virtual Cities by Means of Residents Urban Image: a Wayfinding Support System. In Portugali, J., ed.: Complex Artificial Environments – Simulation, Cognition and VR in the Study and Planning of Cities. Springer, Heidelberg (2006) 245–258 O’Rorke, J., James, G.: Real-time glow. In Gamasutra (May 2004), from http://www.gamasutra. com/view/feature/2107/realtime_glow.php Robinson, A.C.: Highlighting techniques to support geovisualization. Technical report, GeoVISTA Center, Department of Geography, The Pennsylvania State University (2006) Rong, G., Tan, T.S.: Jump flooding in GPU with applications to voronoi diagram and distance transform. In: I3D ’06: Proceedings of the 2006 Symposium on Interactive 3D Graphics and Games, New York, ACM Press (2006) 109–116

210

M. Trapp et al.

Saito, T., Takahashi, T.A.: Comprehensible rendering of 3-d shapes. SIGGRAPH Computer Graphics 24(4) (1990) 197–206 Sester, M.: Application dependent generalization – the case of pedestrian navigation. In: Proceedings of Joint International Symposium on GeoSpatial Theory, Processing and Applications. Vol. 34, Ottawa, Canada, July 8–12 (2002) Trapp, M., Schneider, L., Holz, N., D€ ollner, J.: Strategies for visualizing points-of-interest of 3d virtual environments on mobile devices. In: 6th International Symposium on LBS and TeleCartography, Springer, Berlin, Heidelberg (2009)

Integration of BIM and GIS: The Development of the CityGML GeoBIM Extension Ruben de Laat and León van Berlo

Abstract There is a growing interest in the integration of BIM and GIS. However, most of the research is focused on importing BIM data in GIS applications and vice versa. Real integration of BIM and GIS is using the strong parts of the GIS technology in BIM, and of course the strong parts from BIM technology in GIS. In this paper a mix of strong parts from both worlds is integrated in a single project. The paper describes the development of a CityGML extension called GeoBIM to get semantic IFC data into a GIS context. The conversion of IFC to CityGML (including the GeoBIM extension) is implemented in the open source Building Information Modelserver.

1 Introduction There is an increasing interest in the integration of Building Information Modeling (BIM) and Geospatial Information Systems (GIS) (Akinci et al. 2008; Benner et al. 2005; Clemen and Gr€ undig 2006; Hijazi et al. 2009; IFCwiki.org 2010; Isikdag et al. 2008; Isikdag and Zlatanova 2009b; Wu and Hsieh 2007). A number of publications and projects showed promising results (Clemen and Gr€undig 2006; Hijazi et al. 2009; Isikdag et al. 2008; Nagel 2007; Wu and Hsieh 2007). However, the ‘BIM people’ and the ‘GIS people’ still seem to live in different worlds. They use different technology, standards and syntax descriptions. Previous attempts to integrate BIM and GIS (Hijazi et al. 2009; IFCwiki.org 2010; Kolbe et al. 2005) seem to focus on either BIM or GIS. The two options seen so far are (1) integrating BIM data in the GIS world by using GIS technology, GIS standards and is done by ‘GIS people’ that look at buildings as information in a geospatial context (Benner et al. 2005; Isikdag et al. 2008; Isikdag and Zlatanova 2009b). The other

R. de Laat and L. van Berlo (*) Netherlands Organisation for Applied Scientific Research (TNO), Delft, The Netherlands e-mail: [email protected], [email protected]


211

212

R. de Laat and L. van Berlo

work we see (2) is done by ‘BIM people’ who are modelling advanced detailed 3D buildings with high semantics. They model more buildings including streets; terrain and maybe some underground piping and call this integration of GIS into BIM. Until today the two worlds do not really integrate. BIM is seen as an essential data source for built environments by GIS users (Peachavanish et al. 2006). GIS is seen as a crucial data source for design and integration of new BIM models in a spatial context (Wu and Hsieh 2007). However, while these two worlds are interested in each other’s data, they do not seem to intent to switch in technology or work processes. The authors of this paper see two different worlds that both try to import the other world into their own. There is a need to develop technology to integrate both worlds and create a synergy between the strong (technology) parts of both worlds.

2 Where BIM and GIS Can Learn from Each Other The BIM world and GIS world are quite different. Both worlds have strengths, but both worlds also make progress and first steps in new technologies. A small comparison: The AEC/BIM sector makes intense use of 3D geometry modelled using Industry Foundation Classes (IFC). The ISO standard IFC has a strong focus on constructive solid geometry, boundary representation, Boolean operations, et cetera. The IFC modelled data are mostly file based and exchanged as files (as snapshots of a BIM) by project partners. IFC and BIM usually model buildings and structures above the ground. It is typically used for new buildings and structures. Important concepts in BIM models are the decomposition and specialisation of objects in the model. The relation between objects is of strong importance (Liebich 2009). On the other hand, the GIS world has a server-focused approach. GIS data obviously have a strong focus on the geolocation (using real world coordinates). The relation between geospatial objects is based on the coordinates. The GIS modeller typically models existing data or policies. GIS is strong on 2D geometry and is just starting to experiment with 3D (CityGML 2009; Kolbe 2007; Thurston 2008). We think the BIM and GIS world can create strong synergy. The server approach is getting more and more attention in the BIM world and BIM developers can learn a lot from the experience of the GIS developers. The 3D questions and issues discussed in the GIS world have well known solutions in the BIM world. The BIM and GIS users meet in several complex projects. Both worlds however try to solve the planning questions by using their own technology and way of working. The development and growing use of both CityGML and BIM servers may create a breakthrough in the integration of the two worlds.

Integration of BIM and GIS: The Development of the CityGML GeoBIM Extension

213

3 Integrating BIM and GIS The authors of this paper believe that integrating the two world should be done by using the strengths from both the BIM and GIS world in the context of the other. This means we intent to use a central modelserver for BIM and intense semantics (specialisation, decomposition and relations) and 3D in GIS. To do this, IFC models have to be available online, using a central modelserver (Beetz and Berlo 2010). We have decided to use the open source BIMserver during this project, because it is the only available open source software for this purpose. It also means that the IFC semantics and relations should be available in a GIS context. We have decided to use CityGML for this (Kolbe et al. 2005). It is not possible to integrate IFC semantics into CityGML by default. Therefore we use the extension mechanism for CityGML. A few existing extensions are already available (ADE 2010; Czerwinski et al. 2006). A new CityGML extension will create the possibility to integrate IFC semantics and properties. The open source BIMsever will be able to export IFC data to CityGML, including the IFC geometry, but more important also the semantics and properties. We call the extension on CityGML for IFC data the ‘GeoBIM’ extension (FZK viewer 2010). Of course the integration of BIM and GIS is depending on the assumption that there will be applications from both domains, which can deal with this GeoBIM extension. With the development we try to encourage discussion on this topic.

4 Previous Work on IFC and CityGML Transformation There is a lot of work already done in transforming IFC to CityGML and vice versa (Du and Zlatanova 2006; Hijazi et al. 2009; Isikdag and Zlatanova 2009b). This previous work has a strong focus on converting geometry. IFC geometry uses constructive solid geometry with volumetric, parametric primitives representing the structural components of buildings. 3D GIS (including CityGML) uses boundary representations; accumulation of observable surfaces of topographic features (Fig. 1). This paradigm creates high combinatorial complexity in the transformation. Other pros and cons between IFC and CityGML are described in detail by Isikdag and Zlatanova (2009a). Previous work on matching CityGML and IFC entities, showed the use of semantic information as a priori knowledge and the evaluation of geometrictopological relations between CityGML entities (Isikdag and Zlatanova 2009b). Nagel et al created software to convert IFC geometry to the different levels of detail in CityGML (IFC Engine Series 2009; Nagel et al. 2009). Previous studies on the conversation of IFC to CityGML can be summarized in a few conclusions. First (1) is that previous work is primarily focused on the conversion of geometry. In the conversion of geometry the intention is to convert to different lower Levels of detail (LODs) in CityGML (2). The work so far tries to

214

R. de Laat and L. van Berlo IntBuildingInstallation

IfcWallStandardCase WallSurface FloorSurface

IfcBeam

IfcSlab IfcWindow

Window GroundSurface

InteriorWallSurface

Fig. 1 Geometry modelling paradigms between IFC (left) and 3D GIS (right) (Nagel et al. 2009)

use the rich semantic IFC models to create and feed CityGML models (3) (Isikdag and Zlatanova 2009b). This paper will focus on the extension of CityGML with semantic IFC data. The additional IFC semantics will enrich CityGML in addition to using IFC only as a source for GIS data.

5 Use Cases for a GeoBIM-Extension Since the OGC Web Services testbed phase 4 (OWS-4) the integration of BIM and GIS has some obvious use cases (OWS-4). Targeted application areas explicitly include urban and landscape planning; architectural design; tourist and leisure activities; 3D cadastres; environmental simulations; mobile telecommunications; disaster management; homeland security; vehicle and pedestrian navigation; training simulators; and mobile robotics. The most famous use cases come directly from the OWS-4 testbed. It is the ‘sniper example’ from homeland security. That use-case concentrates on an application where an important politician moves along a particular route. It’s necessary to find all the windows and buildings which have good view on that route and where possibly a sniper can hide. Instead of virtually visiting all building models with a 3D viewer along the route, we might rather want to query the city model to create a report of all corresponding windows, rooms, and buildings in order to check these. Thus we would exploit the semantic information of a city model along the route, and especially the details that come with a highly detailed CityGML or IFC model so we can locate and identify the windows. Because CityGML does not store window width and height it would be very complex to calculate this from the geometry. The window width and height are stored semantically in IFC. These kinds of use cases create the validation for the development of a GeoBIM CityGML extension (Thurston 2008).


215

Other use-cases are: calculating the (indoor and towards the right side of a building) route to critical locations for first responders; locating key structural elements of a building during disasters (IfcStructuralElement); Integrating outdoor navigation software (PNDs) into the indoor domain (IfcStair; IfcRailing; height and width of Door; etc.); evacuation scenario’s for campuses larger than one building; LEED scores for a neighbourhood (instead of just one building); incident simulation and analyses (think of a piping leakage that effects the entrance or exit of buildings. We are aware that some building elements like ‘Stair’ are already in the native CityGML (Stair for example is in the IntBuildingInstallation) but we hope to add value by storing this data more explicit.

6 The Development of the GeoBIM-Extension The development of the GeoBIM extension for CityGML is done on several levels. First the known CityGML object types like Room, Window, Door, Building, etc. are extended with extra properties from IFC. Examples of these properties are the widths and heights of windows and doors. The next level of getting IFC data into CityGML is to extent the ‘AbstractBuiding’ with an extra property what creates a link to the base class of our (to be introduced) extra classes, called VisibleElement. The development has a focus on theoretical possibilities for the transformation of IFC data to CityGML. There is no specific use case to mirror the development. The total IFC schema holds around 900 classes. Most of them are for geometry representation, relations and topology. Theoretical research on what IFC classes could be of use in GIS, showed that there are about 60–70 IFC classes that theoretically could be transformed to a GeoBIM extension (Berlo L van 2009). These classes are listed in Fig. 2.

Fig. 2 List of IFC classes that could be useful in a geospatial context

216


Most of the IFC classes are not of use in a practical GeoBIM use case. For example IfcStructuralPointAction is typically used for structural calculations and therefore has no use in a GeoBIM use case. Applied research has shown that 17 IFC classes are most likely to map to a GeoBIM extension of CityGML. These classes are noted in Fig. 3.

IFC class

CityGML type

Arguments

IfcBuilding

Building

GUID -> GlobalId, Name -> Name

BuildingAddress

Address

-

IfcWall

IfcWindow

IfcDoor

IfcSlab IfcRoof IfcColumn IfcFurnishingElement IfcFlowTerminal IfcColumn IfcSpace IfcStair

InteriorWallSurface or WallGUID -> GlobalId, Surface (Depending on Name -> Name boundaryType) GUID -> GlobalId, Name -> Name, Window OverallWidth -> OverallWidth, OverallHeight -> OverallHeight GUID -> GlobalId, Name -> Name, Door OverallWidth -> OverallWidth, OverallHeight -> OverallHeight RoofSurface or FloorSurface GUID -> GlobalId, (Depending on IfcSlabName -> Name TypeEnum) GUID -> GlobalId, RoofSurface Name -> Name GUID -> GlobalId, Column Name -> Name GUID -> GlobalId, BuildingFurniture Name -> Name GUID -> GlobalId, FlowTerminal Name -> Name GUID -> GlobalId, Column Name -> Name GUID -> GlobalId, Room Name -> Name GUID -> GlobalId, Stair Name -> Name, ShapeType -> Type

IfcRailing

Railing

IfcAnnotation

Annotation

IfcColumn

Column

IfcBeam

Beam

GUID -> GlobalId, Name -> Name, PredefinedType -> PredefinedType GUID -> GlobalId, Name -> Name GUID -> GlobalId, Name -> Name GUID -> GlobalId, Name -> Name

Fig. 3 Mapping of IFC classes to CityGML types; including arguments and attributes


217

The properties of these classes can also be transformed to CityGML attributes. Some classes from IFC map direct to a corresponding CityGML type. For example IfcBuilding maps directly to Building in CityGML. Other mappings are represented in Fig. 3. The result of the development of the GeoBIM extension (ADE) for CityGML is presented in an XML Schema file (XSD). The result is also represented as a UML class diagram shown in Fig. 4. All added properties from IFC are presented in the CityGML file. The GeoBIM extension creates some new objects in CityGML. An example of such a new object type is ‘Stair’. This object has some properties and also has geometry.

7 Prototype Implementation of the GeoBIM-Extension To create a practical use, the GeoBIM extension is implemented in the open source Building Information Modelserver (BIMserver) (BIMserver 2009). The software implementation of the transformation from IFC to GeoBIM in the open source BIMserver creates a situation where the theoretical model will be tested by implementers. Both the theoretical model and the software implementation feed each other with experience and results. This makes the theoretical extension very robust for practical use. The open source BIMserver architecture consists of an EMF model (EMF 2010) of IFC, a BerkeleyDB database (2010) and several interfaces for communication (REST, SOAP, webuserinterface). The open source BIMserver is intended to be a tool to support innovative collaboration in the AEC sector. The storage of BIM information is native IFC. Key features of the open source BIMserver are the ability to merge and query IFC models (Beetz and Berlo 2010). For this reason the BIMserver software does not need the ability to compose and calculate complex geometry of IFC. However, this feature is needed to transform IFC geometry to CityGML geometry. For this we connected the IFC Engine DLL library (IFC Engine Series 2009) to the EMF interface. Furthermore, the CityGML4j java library (CityGML4j 2010) is used to compose CityGML files. This CityGML4j is also connected to the EMF interface of the BIMserver software (Fig. 5). The conversion of IFC data to CityGML is done on object level of IFC data. The steps that are taken: l l l l l l

Get an object from IFC (BIMserver) Run the object through the IFC Engine DLL (IFC Engine DLL) Get triangles from the object (IFC Engine DLL to BIMserver EMF interface) Get IFC properties belonging to the object (BIMserver EMF core) Get next object (BIMserver EMF core) Convert data from memory to CityGML file (CityGML4j)

Fig. 4 The GeoBIM extension (ADE) for CityGML represented as a UML Class diagram. Note: this is not the UML schema of the complete CityGML schema including the new extension; this is the schema of only the extension XSD

218 R. de Laat and L. van Berlo


219

Fig. 5 The schematic representation of the open source BIMserver software architecture (Inspired from BIMserver and the potential of server side BIM 2009)

The IFC objects with an equal object in CityGML/GeoBIM (for example IfcDoor and Door or IfcWindow and Window) will be converted to the correct CityGML objects. These objects in CityGML get the extra properties from IFC.

8 Prototype Testing During and after the development of the ADE and the implementation in the BIMserver, we tested the conversion. During the testing, three publicly available IFC files were used (IFCwiki.org 2010). First result of this test is the notice that some viewers (like the widely used LandExplorer from Autodesk (LandExplorer 2010) do not display geometry of objects defined in an extension. This means that no stairs (and other semantically added objects) are shown in (for example) LandExplorer. Other viewers (like the FZK viewer 2010) do not have this issue and show the result just fine (Figs. 6 and 7).

220


Fig. 6 Result of a conversion from IFC to CityGML including the GeoBIM extension. The added properties are to be seen in the properties view in CityGML. At the top the original IFC file. It is clear to see that the geometry of the stairs and fence is not visible in the CityGML result. Viewing the same CityGML file in the FZK viewer does show the geometry of both the stairs and the fence (small right)

After conversion of the IFC to CityGML (including the extension data) the size of the files increased by a tenfold or more. File number 1 is 4.6 MB as STEP IFC file and 114.7 MB as CityGML file (about 25 times as big). File number 2 is 2.8 MB as STEP IFC file and 106.1 MB as CityGML file (about 38 times as big). File number 3 is 2.9 MB as STEP IFC and 31.6 MB as CityGML (about 11 times as big) (IFCwiki.org 2010). We have to remark that this is with very basic geometry representation using triangles. It is a known issue that representation in XML and especially GML is data intensive. The optimization of geometry transformation between IFC and CityGML will drop the CityGML file sizes. Using Gzip or any other ZIP protocol can also solve this problem. We implemented a ‘download as ZIP’ option in the open source BIMserver to decrease the resulting CityGML file sizes. During the development of the GeoBIM extension to get IFC data into CityGML the authors found no possibility to semantically create a network structure in CityGML. For ring piping for example, the final pipe cannot be (semantically) connected to the first. In the AEC sector this is a much-used method. For example heating systems and sewerage make intensive use of ring piping. Getting this semantic information into CityGML is a key issue for the link between BIM and GIS. Recent studies show that the next version of CityGML could make this possible. The final issue we found during development and testing is the freedom IFC gives to users (e.g. software implementers) to represent data. In pure form IFC


221

Fig. 7 Another result from the IFC to CityGML conversion including GeoBIM extension. On the top right the original IFC file. On the top left the result of the hallway in Autodesk LandExplorer. The geometry of the stairs is missing. Viewing the same CityGML file in FZK viewer does show the geometry of the stairs

makes use of just a few base objects. All other objects are specializations of these base objects (IFCwiki.org 2010). This means there is no single way to connect a specific IFC object with another. For example: An IfcWindow can be connected to an IfcOpening, which is connected to an IfcWall, which is connected to an IfcSpace, which is connected to an IfcBuilding. This route to find out which Window is connected to which Space (Room in CityGML) and Building is a chosen route in a specific data file, but not a statically defined one on IFC schema level. The link could also be IfcWindow – IfcBuildingStorey – IFC Space (for example). These kinds of links are much more statically defined in CityGML. A connection between a Window, Wall and Room in CityGML is always the same. This makes it very difficult for software implementations to transform IFC data into CityGML data. This problem is inherent to the IFC schema structure and will probably not be solved. This issue is also a reason why not all 60–70 semantic objects and their properties will be present in the GeoBIM extension. In theory this would be possible, but in practice no software can fully implement this transformation. All information, object, properties and relations from IFC that are stored in the GeoBIM extension are available in the generated CityGML exports. The use of this information in CityGML is very welcome to both BIM and GIS users (Benner et al. 2005; Borrmann and Rank 2009).

222


9 Discussion The results presented in this paper could help to integrate BIM and GIS. However, some elements still need some discussion and remarks. First of all, the geometry issues known in the transformation from IFC to CityGML (Nagel et al. 2009) are still not solved. This work does not contribute to a better transformation and only used basic triangles for the geometry in CityGML. This is also why the file sizes of CityGML files are between 11 and 38 times as big. The solution to this issue is needed to get practical use of this integration. For now the implementation in the open source BIMserver only exports IFC to CityGML LOD4, including the GeoBIM extension data. To use IFC to CityGML transformation in practice, the transformation to lower LODs is necessary. This work is already done but implemented in closed source commercial software (Explorer 2008). IFC data are used to exchange information in the AEC sector. It is modelled for this purpose and therefore the use of textures in IFC is rare. Almost all the transformations from IFC to CityGML will be without textures. The growth of semantic data in CityGML and the growing complexity of 3D geometry representations might cause a situation where the usability of the data decreases. Of course the most common and most valuable argument for this issue is that CityGML was never designed, nor intended to be used for these applications. To prevent this from happening we think of splitting 3D geometry representation and semantic information. The 3D geometry is nothing more than one of the properties of an object. It is not necessary to serve this property in all use-cases. Another possible option is to start defining 3D geometry by using binary models and standards. The use of ‘human readable’ XML lowers performance and usability of 3D models on the web. The use of binary standards might help to speed up the adaptation of 3D usage on both the Geoweb and the BIMweb. When CityGML extensions with geometry representation are not shown in (some) viewers, it is not clear how practical use-cases will develop in the future. Software developers of CityGML viewers should extend their software to view geometry representations of objects in a CityGML ADE extension.

10

Conclusions

This study investigates the integration of BIM and GIS. Main part of the research was the development and implementation of a GeoBIM extension on CityGML for IFC data. To fully integrate BIM and GIS it is obvious that a translation from CityGML to IFC is also necessary. Since this paper is describing a development, which is not finished, there are no real conclusions, which can be made. However, some first findings and conclusions can be stated.


223

So far we concluded that it is technically possible to add semantic information from IFC into CityGML using the developed GeoBIM extension. The GeoBIM extension works in practice and is implemented in software. The conversion of IFC data to CityGML files with additional rich IFC semantics is proved to be possible. Both IFC and CityGML have made decisions during the design and development of their native schemas that impose restrictions on the integration. Both IFC and CityGML are about to change their schema definitions. IFC to 2 4 will be a more strict definition, and CityGML 1.1 will probably have the ability to create network structures. We are aware that CityGML was not originally designed and implemented to be 100% consistent with the semantics and content available in an IFC. CityGML was originally designed for sharing 3D city models and not internal infrastructure, such as piping systems. The conclusion for now is that it will be difficult to get full IFC semantics in CityGML, and we are aware of the question if CityGML should be overloaded with additional extensions for which it was not originally designed, but in the future the technical possibilities will increase. To fully integrate BIM and GIS the AEC sector needs to start working with central modelservers. The use of central servers in the BIM world is still to be adopted. This experiment had a strong focus on transforming IFC semantics into CityGML and showed promising results.

11

Future Work and Ambitions

In the future the GeoBIM extension should be updated given the new possibilities created by the new releases of both the IFC and CityGML schemas. The ambition of the team is to implement the conversion of IFC to CityGML not only for the LOD 4 (including the GeoBIM extension) but for the LODs 0–3 as well. Another ambition is to implement an interface to spatial query building information models (Borrmann and Rank 2009) in the open source BIMserver, using the CityGML GeoBIM extension. Future work should also focus on testing the use, investigate the benefits and results in practice. Since only lab testing has been done during this research, there is a need for more use-case testing on the practical implications of this technology. To fully integrate BIM and GIS a translation from CityGML to IFC is a main issue that should be investigated and developed.

References Akinci B, Karimi H, Pradhan A, Wu CC, Fichtl G (2008) CAD and GIS interoperability through semantic web services. ITcon 13:39–55 Beetz J, Berlo L. van (2010) Towards an Open Building Information Model Server. DDSS 2010 Benner J, Geiger A, Leinemann K (2005) Flexible generation of semantic 3D building models. In: Gr€oger G, Kolbe T (eds) Proceedings of the 1st International Workshop on Next Generation 3D City Models, Bonn, pp. 17–22

224


Berkeley DB (2010) open source database. Retrieved August 1, 2010 from the World Wide Web: http://www.oracle.com/technetwork/database/berkeleydb/overview/index.html Berlo L van (2009) “CityGML Extension for BIM/ IFC information”, presented at the Free and open source for GIS conference FOSS4G, October 2009 BIMserver and the potential of server side BIM (2009) Retrieved February 2009 from the World Wide Web: http://www.stress-free.co.nz/bimserver_and_the_potential_of_serverside_bim BIMserver (2009) Building information model server. Retrieved November 2009 from the World Wide Web: www.bimserver.org Borrmann A, Rank E (2009) Specification and implementation of directional operators in a 3D spatial query language for building information models. Advanced Engineering Informatics 23:32–44 CityGML (2009) CityGML Encoding Standard document version 1.0.0. Retrieved April 20, 2010 from the World Wide Web: http://www.citygml.org/1522/ CityGML4j (2010) Java class library and API for facilitating work with CityGML. Retrieved August 1, 2010 from the World Wide Web: http://opportunity.bv.tu-berlin.de/software/ projects/show/citygml4j/ CityGML – ADE (2010) CityGML application domain extensions. Retrieved April 20, 2010 from the World Wide Web: http://www.citygmlwiki.org/index.php/CityGML-ADEs Clemen C, Gr€undig L (2006) The industry foundation classes-ready for indoor cadastre? In: Proceedings of XXIII International FIG Congress, Engineering Surveys for Construction Works II, Munich, Germany Czerwinski A, Kolbe T, Pl€ umer L, St€ ocker-Meier E (2006) Interoperability and accuracy requirements for EU environmental noise mapping. In: Kremers H (Hg.) Proceedings, InterCarto – InterGIS 12, Berlin Du Y, Zlatanova S (2006) An approach for 3D visualization of pipelines. In: Abdul-Rahman A, Zlatanova S, Coors V (eds) Innovation in 3D-Geo Information System, Springer, Berlin, Heidelberg, pp. 395–404 EMF (2010) Eclipse Modelling framework. Retrieved April 20, 2010 from the World Wide Web: http://www.eclipse.org/modeling/emf/ FZK viewer (2010) Tool for viewing IFC files. Retrieved April 20, 2010 from the World Wide Web: http://www.iai.fzk.de/www-extern/index.php?id¼1134 Hijazi I, Ehlers M, Zlatanova S, Isikdag U (2009) IFC to CityGML transformation framework for geo-analysis: a water utility network case. In: Maeyer P de, Neutens T, Rijck M de (eds) 3D GeoInfo, Proceedings of the 4th International Workshop on 3D Geo-Information, Ghent University, Ghent, pp. 123–127 IFC Engine Series (2009) Library for handling IFC models. Retrieved April, 20, 2010 from the World Wide Web: http://www.ifcbrowser.com/ IFC Explorer (2008) Tool for viewing and conversion of IFC models. Retrieved April, 20, 2010 from the World Wide Web: http://www.iai.fzk.de/www-extern/index.php?id¼1566 IFCwiki.org (2010) Website that hosts information about Industry Foundation Classes. Retrieved October 1, 2010 from the World Wide Web: http://www.ifcwiki.org Isikdag U, Underwood J, Aouad G (2008) An investigation into the applicability of building information models in geospatial environment in support of site selection and fire response management processes. Advanced Engineering Informatics 22:504–519 Isikdag U, Zlatanova S (2009) A SWOT analysis on the implementation of Building Information Models within the Geospatial Environment. In: Krek A, Rumor M, Zlatanova S, Fendel EM (eds) Urban and Regional Data Management – UDMS Annual 2009, Taylor & Francis Group, London, pp. 15–30 Isikdag U, Zlatanova S (2009) Towards defining a framework for automatic generation of buildings in CityGML using building Information Models. In: Lee J, Zlatanova S (eds) 3D Geoinformation and Sciences, Springer, Berlin, Heidelberg, pp. 79–96 Kolbe T (January 2007) 3D Geospatial Information Modelling with CityGML. Presentation for the OGC on January 18, 2007


225

Kolbe, T, Gr€oger, G, Pl€ umer, L (2005) CityGML – Interoperable Access to 3D CityModels. In: Proceedings of the International Symposium on Geo-information for Desaster Management on 21–23 March 2005, Delft LandExplorer (2010) Tool for viewing CityGML files. Retrieved April 20, 2010 from the World Wide Web: http://www.3dgeo.de/citygml.aspx Liebich T (2009) IFC 2x Edition 3 Model Implementation Guide v1.7. Retrieved April 20, 2010 from the World Wide Web: http://www.iai tech.org/downloads/accompanyingdocments/ guidelines/IFC2x%20Model%20Implementation%20Guide%20V2 0b.pdf Nagel C (2007) Conversion of IFC to CityGML; Meeting of the OGC 3DIM Working Group at OGC TC/PC Meeting, Paris (France) Nagel C, Stadler, A, Kolbe T (2009) Conceptual Requirements for the Automatic Reconstruction of Building Information Models from Uninterpreted 3D Models, Academic Track of Geoweb 2009 Conference, Vancouver OWS-4, OGC Webservices, phase 4; http://www.opengeospatial.org/projects/initiatives/ows-4, last accessed [01-2011] Peachavanish R, Karimi H, Akinci B, Boukamp F (2006) An ontological engineering approach for integrating CAD and GIS in support of infrastructure management. Advanced Engineering Informatics 20(4):71–88 Thurston J (June 2008) Interview: CityGML – Modeling the city for the future. Retrieved August 28, 2009 from the World Wide Web: http://www.vector1media.com/index2.php?option= com_content&do_pdf=1&id=3251 Wu I, Hsieh S (2007) Transformation from IFC data model to GML data model: Methodology and tool development Journal of the Chinese Institute of Engineers 30(6):1085–1090

.

Modelling Three-Dimensional Geoscientific Datasets with the Discrete Voronoi Diagram Tom van der Putte and Hugo Ledoux

Abstract Geoscientific datasets are often formed by scattered samples in 3D space having highly anisotropic distribution. To model the continuity of the phenomena they represent (e.g. temperature of a body of water, or percentage of a chemical in the air) raster structures are in most cases used. To overcome the shortcomings of rasters the Voronoi diagram (VD) has been proposed as an alternative. However, while in theory the VD is a sound solution, its use in practice is hindered by the fact that it is complex to construct and to manipulate (removal of samples, interpolation, etc.), and spatial tools have to be built. We propose in this paper a ‘middle’ solution: the 3D discrete Voronoi diagram (DVD). We investigate the properties of the 3D DVD, we propose algorithms to construct and manipulate it, and we demonstrate its use in practice with a prototype that we have built. Our prototype uses existing tools for visualisation and further analysis of DVDs.

1 Introduction The geoscientific disciplines such as oceanography, meteorology and geophysics are different from the other applications of 3D GIS because, instead of modelling man-made objects (e.g. houses, bridges, tunnels), we have to model the spatial distribution of continuous geographical phenomena in 3D space. Examples of these phenomena are the salinity of a body of water, the humidity of the air or the percentage of a certain chemical in the soil. The representation and analysis of such phenomena is complicated by the fact that the collection of samples is problematic: it is often impossible to measure everywhere these phenomena and as a results the This work was carried out while the first author was a student in the M.Sc programme Geographical Information Management and Applications (GIMA). T. van der Putte (*) TNO, Geological Survey of the Netherlands Princetonlaan 6, Utrecht 3584 CB, The Netherlands e-mail: [email protected] H. Ledoux GIS Technology Group, Delft University of Technology, Delft 2628 BX, The Netherlands e-mail: [email protected]


227

228

T. van der Putte and H. Ledoux

datasets are formed by sparse and anisotropically distributed samples in 3D space. Three-dimensional continuous geo-information is usually modelled with raster structures, either directly with 3D grids (voxels) or with hierarchical grids such as octrees (Bak and Mill 1989; Jones 1989). This is due to the fact that raster structures are simple structures naturally stored and manipulated in a computer (with arrays of numbers) and hence a certain number of modelling tools are available. However, raster structures also have several disadvantages both in theory and in practice (Fisher 1997; Kemp 1993): (1) space is arbitrarily divided without taking into account the original samples; (2) the size of a 3D raster can become huge is a fine resolution is wanted; (3) rasters scale badly and are not rotationally invariant; (4) the original samples are ‘lost’. To circumvent these problems, Ledoux (2006) and Ledoux and Gold (2008) have proposed the Voronoi diagram (VD) as an alternative and showed that it is useful not only for the representation of 3D phenomena, but also for their analysis. The main advantages are: (1) the tessellation of the 3D space obtained with the VD gives a clear and consistent definition of neighbourhood of the samples and adapts to the distribution of these; (2) the continuity of the 3D phenomena can be reconstructed with Voronoi-based interpolation methods; (3) the structure is locally modifiable, which permits us to interactively explore a dataset and manipulate it; (4) it enables several visualisation operations, as well as several spatial analysis operations. While the 3D VD is conceptually superior to raster structures, it has drawbacks in practice. It is indeed rather difficult to construct it in a robust and efficient way (Field 1986; Sugihara and Inagaki 1995), and manipulation algorithms (movement and removal of points) are problematic to implement and as far as we know only exist for the VD of points in Euclidean space (Devillers and Teillaud 2003; Ledoux et al. 2005; Hoffmann 1989; Russel 2007). Another obstacle to its use by practitioners is that specialised tools have to be built. In this paper we investigate the use of the 3D discrete Voronoi diagram (DVD) for the modelling of geoscientific datasets. The DVD is one natural variant of the ‘normal’ VD; several other variations are possible, see ? for an exhaustive list. It can be seen as a ‘middle’ solution between the VD and raster structures. Both models have pros and cons, and it is interesting to study which ones will be retained with the DVD. As Fig. 1a shows, given a set of points in space the VD divides the space into cells (called Voronoi cells) in such a way that every location within each cell is closest to the points that lies within that cell compared to all other points; the boundaries of the cells represent the locations that are equidistant from two or more points. The DVD of the same pointset is shown in Fig. 1b. With it the cells are formed by groups of pixels having the same ID (here a colour) and we can notice that they have a similar shape; but when we zoom-in on an edge between two cells, the tessellation becomes visible. The DVD is also possible in 3D: the cells are convex polyhedra formed by groups of voxels. We describe further in Sect. 2 the DVD and discuss related work. We also present in Sect. 3 new algorithms to construct and manipulate (e.g. removal of points) 3D DVDs. Observe that we are particularly interested in dynamic solutions where users can modify interactively DVDs (as in Anselin 1999) because we believe it to greatly help the exploration and

Modelling Three-Dimensional Geoscientific Datasets with the Discrete Voronoi Diagram

229

Fig. 1 (a) The VD for a set of points in 2D. (b) The DVD in 2D for the same set of points; the zooming in on a small section of the DVD shows the discrete nature of the structure

understanding of complex 3D datasets. We have implemented these algorithms and used them for the modelling of geoscientific datasets. We describe briefly in Sect. 4 the architecture we used, which is a mix of our own code and the use of the opensource tools for the handling, analysis and visualisation of the 3D DVDs.

2 Work Related to the 3D DVD Given a set of points in 3D space (also called ‘seeds’), the 3D VD divides the space into convex polyhedra that represent the ‘closest’ space around each seed. As Fig. 2 illustrates, for the 3D DVD of the seeds the space is tessellated into regular cells (cubes) that we call voxels. The Voronoi cells in discrete space are represented by a group voxels which share the same value (the ID of the seed generating the Voronoi cell). It should be noticed that while the Voronoi cells of the seeds on the boundary of the convex hull are in theory unbounded, for the DVD they have to be bounded arbitrarily (as the Fig. 2 shows). Constructing the DVD for a set of seeds therefore boils down to finding the closest seed of every voxel and assigning that ID. When a voxel is situated at exactly the same distance from two or more seeds we have a ‘tie’ or a ‘conflict situation’. Notice that in this paper we use Euclidean distances and not a different metric (e.g. the Manhanttan distance). The distance between two voxels is thus a Euclidean distance between the two centres of the voxels. However, using another metric would be relatively simple and could help building generalised Voronoi diagrams (Okabe et al. 1994); we discuss in Sect. 5 the possibilities. We review in this section the algorithms and methods that have been proposed to construct a DVD. It should first be noticed that of all methods that we discuss only one was specifically developed for the 3D DVD, the rest were aimed towards the 2D VD; the generalisation of these is however technically possible.

230


Fig. 2 Left: an example of a 3D DVD, only the boundary of the space is shown. Right: a slice from the DVD, which helps understanding the spatial distribution of the points in space

Fig. 3 The implicit and explicit methods to construct a DVD

To construct a 3D DVD, we distinguish between two different types of methods (as shown in Fig. 3): l

Explicit methods calculate the value to assign to each voxel by finding the closest seed in the input dataset.

Modelling Three-Dimensional Geoscientific Datasets with the Discrete Voronoi Diagram l

231

Implicit methods assign a voxel to a seed based on the value of the voxel next to them; several passes over a grid is needed. They are theoretically efficient since distance calculations are not required for every voxel. However, a control mechanism is imperative to ensure that voxels are assigned correctly.

2.1

Explicit Methods

A brute-force implementation of the explicit method with n seeds and m voxels would require for all m voxels to calculate the distance to each seed and assigning the ID of the closest, which is O(n m) – with an arguably small tessellation of 100 100 100 voxels with only 500 seeds, 500 millions distances have to be calculated. A major disadvantage of this method is when trying to add, move or remove seeds: it requires the entire VD to be rebuilt. The brute-force algorithm can however be sped up with the use of auxiliary data structures, such as the kd-tree. Park et al. (2006) use it to create both the 2D and the 3D DVD. A kd-tree recursively partitions the space into a binary tree that can be used to efficiently solve the nearest neighbour problem. However, the efficiency of nearest neighbour queries requires a balanced tree and after points have been inserted or removed the tree becomes unbalanced and has to be re-balanced. But the tree structure and the DVD are not dynamically linked; in other words, when updating the tree, the DVD is not updated automatically. The main advantage of explicit methods is that they are not error-prone. Several people have devised ways to create a DVD with the aid of the GPU (graphical processing unit) instead of the CPU, see among others Hoff et al. (1999) and Rong and Tan (2006, 2007). While these methods are very fast for the 2D case, they do not generalise directly to the 3D case, except Rong and Tan (2007) who propose a slice-by-slice approach. Another issue with GPU-based methods is that an approximation of the DVD is sought for visualisation purposes (i.e. they do not guarantee that a correct DVD is created in case of ties), and that could be problematic if this DVD is used for further analysis.

2.2

Implicit Methods

Schueller (2007) proposes a 2D method based on the principle of expanding circles around each seed to assign IDs (principle of dilation). It is theoretically more efficient than the brute-force explicit algorithm because only the pixels at the boundaries between two or more Voronoi cells have to be checked to determine the closest seed. In other words, as long as circles of two seeds do not touch, all the pixels within that circle are correctly assigned to the corresponding seed. To assess which pixels are neighbours of a claimed pixel P, he proposes different neighbourhoods: the Moore and the Neumann neighbourhood (also called in 2D the 8- and the 4-neighbourhood), as shown in Fig. 4a. These are approximations of a circle and are

232


Fig. 4 (a) Left: the Moore neighbour. Right: the Neumann neighbourhood. (b) Three rounds of dilation for the Neumann neighbourhood

dilated around each seed; Fig. 4b shows the first three rounds of dilating a Neumann neighbourhood. Li et al. (1999) use the dilation principle to create the DVD of different shapes (lines and polygons) but to make their implementation faster they use the Manhattann instead of the Euclidean distance. Zhao et al. (2002) also use the dilation concept but in combination with a quadtree. They reason that when dilating seeds into Voronoi cells, it is only necessary to calculate distances between a pixel and two or more seeds for pixels near the boundaries of Voronoi cells. Pixels that are closer to the seeds might be assigned to a seed more effectively. This is done by creating a quadtree of the raster, so that whole areas of pixels that belong to a certain seed can be assigned that value in one operation.

3 Our Algorithms to Construct and Manipulate a 3D DVD We describe in this section the algorithms that we have designed to construct a 3D DVD, to add/remove a single point to/from a 3D DVD, and to convert one DVD into a continuous raster with natural neighbour interpolation. The algorithms we propose are based on the dilation method of Schueller (2007); we generalised it to 3D and we made several modifications to the dilation component to make it more efficient. First observe that in 3D the Moore neighbourhood consists of a 3 3 voxels cube, and that the Neumann neighbourhood consists of a 6-voxel object, where all six faces of the central voxel are adjacent to one of the six voxels. These approximate a sphere in raster space, but, because they are not perfect spheres, when dilating them errors can occur. The challenge is ensuring that no errors occur while minimising the number of distances calculated. Figure 5 illustrates the problem with a 10 7 pixels in which two seeds have been inserted. These seeds are iteratively dilated with a Moore neighbourhood structuring element, resulting in expanding squares. The green voxel in the figure shows a situation that could introduce errors. In the last round, this has been claimed by the blue seed, but when looking purely at the Euclidean distance, the distance to the blue seed is larger than that of the orange. The way to overcome this is to allow the algorithm to reassign voxels that have already been claimed.


233

Fig. 5 Dilation of two seeds and the incorrect claiming of a pixel/voxel

Notice that because of space constraints we cannot give all the details of the algorithms but these are available – including pseudo-code – in the Master’s thesis of the first author (van der Putte 2009). A proof of correctness for the construction, and its termination, is also available.

3.1

Construction of the 3D DVD

The following construction algorithm assumes that all seeds are known in advance and construct the DVD in one operation. The general idea of the algorithm is as follows. First all the voxels containing an input seed are assigned an ID, and the other ones are assigned NoData. Then a Neumann neighbour is used (we have used and tried others but the Neumann gave the best result, as it did for Schueller 2007) and placed over every voxel that has been given a ID value already, which actually ‘stamps’ the structuring element onto the original 3D grid. The algorithm continues until every voxel has been assigned an ID (the termination condition is that no voxel has been modified during one pass). At each step of the algorithm the voxels with IDs are dilated, which means that they try to claim their neighbours. If the neighbour has the same ID value as the current voxel v, it is correctly assigned already; if a neighbour has NoData it is automatically assigned the ID of v; if an ID has been previously set, but not the same as that of v, then we have a conflict. The Euclidean distance to both claiming seeds are calculated and the minimum wins. If the two distances are exactly the same an arbitrary choice must be made, which can be based for instance on the lexicographic order of the IDs or first-come-first-served. Notice that the rule can influence significantly the resulting DVD, as Fig. 6 shows. Observe that the only voxels that need to be dilated are in fact the voxels on the boundary of each dilating area; those ‘inside’ a dilating Voronoi cells have already been processed. This criterion can be narrowed down even further by stating that only the boundary voxels of Voronoi cells that are still dilating need to be processed at each step of the algorithm. This removes the need for dilating those voxels that are at boundaries that are stationary, either at the edges of the image or at the stable boundaries between areas. In effect, this means that only the voxels that – in the current pass – have been assigned a different ID value than they originally carried need to be considered for dilation. The improvement to the algorithm is done by keeping a list of these voxels, and since we use arrays we have direct access to them.

234


Fig. 6 Random assignment of two equidistant seeds. The pixels/voxels of the middle column could be assigned to either seeds, which would yield different DVDs

Fig. 7 The process of removing a seed and its Voronoi cell. (a) Three Voronoi cells. (b) Identifying the green voxels and boundary voxels. (c) The darker coloured boundary voxels are then redilated until all green voxels are overwritten

3.2

Insertion and Removal of Points

The same mechanism that enables a seed to propagate throughout a grid to all ‘correct’ voxels in the construction algorithm can be readily used to insert a new seed in a valid DVD. To do this, first the voxel representing the new seed has to be assigned the new ID, and the list of voxels to process contains only that seed. Then, the neighbours will be reassigned if they are closer to the new seed then the old ones. In this way, the new DVD will be finished in as many rounds as it takes to dilate the new seed. This method will not only allow for the insertion of one point, but can be used to simultaneously insert multiple new points. Building the adapted DVD will then take only as many rounds as are necessary for the slowest Voronoi cell to be dilated. Incremental construction of the DVD is thus possible without designing new algorithms. In vector space, removing a point from a 3D VD is problematic when degenerate cases are present (Devillers and Teillaud 2003; Ledoux et al. 2005). By contrast, in raster environment it is relatively straightforward and there are no degenerate cases. The idea behind the algorithm is depicted in Fig. 7 and is as follows. First, all the


235

voxels having the ID to be deleted must be identified. We start at the seed to delete and expands until we reach the border of the cell, and we keep a list of the boundary voxels. Second, these voxels are processed with the same dilate function as is used for the incremental insertion; that will ‘re-dilate’ the Voronoi cell surrounding the removed Voronoi cell, overwriting the voxels that are to be deleted, finally yielding a recreated Voronoi diagram (Fig. 7c).

3.3

Natural Neighbour Interpolation

One direct possible use of the DVD is to perform a natural neighbour interpolation on the input dataset. The natural neighbour interpolation is a method that can be performed with the VD or the Delaunay triangulation as a basis (Sibson 1981), and is used in several fields that make use of datasets containing scattered data such as engineering, computer sciences and geosciences (Gold 1989; Watson 1992; Sambridge et al. 1995). Although the concepts behind natural neighbour interpolation are simple and easy to understand, its implementation is far from being straightforward, especially in higher dimensions (Owen 1992; Perez and Traversoni 1996; Ledoux 2006). The main reasons are that the method requires the computation of two VDs – one with and one without the interpolation point – and also the computation of volumes of Voronoi cells. While these operations are error-prone in vector space, we have seen that in raster space they are simple and degenerate cases do not arise. Moreover, the volumetric calculations that can be difficult to implement in vector space are replaced by simply counting voxels, and multiplying the number of voxels by their (known) volume, as Fig. 8 shows.

Fig. 8 Interpolation a DVD using natural neighbour interpolation. (a) The DVD. (b) The inserted point q, to be interpolated, and the corresponding new Voronoi cell (grey area). (c) The amount of voxels within the black outline corresponding to a specific seed, relative to the total amount of voxels within the outline, determines the weight for each of the corresponding seeds

236


4 Modelling Continuous Phenomena in Geoscience To test our approach for the modelling of 3D geoscientific datasets, we have implemented a prototype whose architecture is described in Fig. 9. We have implemented all the algorithms described in the previous section with the Python language (using the well-known numpy package for n-dimensional arrays). To be able to analyse and visualise the 3D DVDs we use GRASS,1 the only off-the-shelf GIS that permits us to handle 3D grids (as far as we know). Other GIS packages do handle 3D grids such as NetCDF, but they slice them to obtain 2D grids before carrying out analysis. For the visualisation of 3D grids, we found GRASS’s capabilities limited and instead decided to use MayaVi2.2 It is an implementation of the Visualization Toolkit (VTK)3 and permits us to import different 3D formats (both vector and grids) and to perform several visualisation operations. While our solution does involve exporting and importing DVDs in different formats (we also wrote some conversion scripts), it has the benefits of using well-known tools and no new software has to be developed from scratch. To show the possibilities the 3D DVD offers some examples of its usage are shown here. The dataset used is a geological dataset containing the concentration of a chemical substance; it contains 150 data points, each with (x, y, z, attribute). Normally a point dataset would be much larger, ranging from 10,000 to 100,000 points, however for this showcase a small dataset is used to be able to show details clearly. In Fig. 10 the distribution of the sample points in this test dataset is shown. It can be seen that most of the points are taken along lines parallel to the z-axis (blue), which indicates that the data is taken from for instance drilled wells. The anisotropy of the distribution of the sample points is also clear from this image. The 3D DVD of the dataset is shown in Fig. 2, and all the figures from this section were made with that dataset. The 3D DVD, when combined with a GIS and a visualisation tool, can be used in different ways; what follows is an overview.

4.1

To Establish Neighbourhood Relations Between Unconnected Points

This is especially useful for anisotropically distributed datasets, because the VD permits us to find points that are both close to a location, and located ‘around’ that location. There is not much difference in the way the exact and the DVD handle these relations. Adjacency is in this case the key, which, for the exact version, is 1

http://www.grass.itc.it http://www.code.enthought.com/projects/mayavi/ 3 http://www.vtk.org 2


237

Fig. 9 Schematics showing the different software packages and the accompanying data formats

Fig. 10 Distribution of points in the test dataset. The anisotropy in distribution is clearly visible

usually explicitly stored. In the discrete representation of the 3D VD, although not implemented yet, storing adjacency would be easily added as a functionality – this would also permit us to derive the Delaunay tetrahedralization from a DVD.

238

4.2


To Determine the Area of Influence of Points

For the exact 3D VD, a Voronoi cell (which represents the area of influence) is described by a polyhedron, whereas in the DVD it is determined by the collection of voxels that carry the same ID. This means that it is relatively easy to determine the area of influence by simply finding all voxels that carry the same value. In this way sub-selections can easily be made and be used in combination with different spatial analysis operations, such as map algebra. Reclassification is one of these operations. Suppose a user wants to divide the attribute values into three classes (low, medium and high values). This can be easily done using 3D map algebra functionalities (Tomlin 1983; Mennis et al. 2005), as implemented in GRASS. Figure 11 shows a cut plane through a reclassified 3D DVD (the reclassification is based here on the value of the attribute).

4.3

To Support Numerous Visualisation and Spatial Analysis Operations

The exact 3D VD is difficult to visualise and analyse, and special tools have to be built (Ledoux and Gold 2008). By contrast, the 3D DVD can be used with off-theshelf tools. Figure 12 shows that by slicing the data the distribution of points can be easily understood by the user. With tools such as MayaVi2, this plane can be controlled interactively in any directions. It must be noted here that the resulting slices do not represent a 2D VD. As for the spatial analysis operations, it has already been mentioned that with the DVD Voronoi cells are easily identified, and since the raster format is used,

Fig. 11 A cut plane through a reclassified 3D DVD, showing a reclassification into three classes


239

Fig. 12 Visualisation of a DVD by using a special slicing tool in MayaVi2, allowing for 3D visualization control. Three consecutive slices of one DVD are shown

performing spatial analyses on these structures is relatively easy. This is one of the points that the strength of the DVD may lie in, since spatial analysis operations for grids are simple, well-known and already implemented in GRASS. One example is the creation of a mask for a given grid if one wants to show only a particular section of the raster, based on the value of each voxel. Another example is the creation of neighbourhood (local) filters. To create these filters, a function is provided that enables the user to incorporate values of neighbouring voxels for each evaluated voxel. Using this function, it is possible create for instance raster representing the derivative of a continuous field (like the slope for grids of elevation).

4.4

As a Prerequisite for Natural Neighbour Interpolation

As explained in Sect. 3.3, the DVD is well-suited for recreating a 3D grid with the natural neighbour interpolation. Figure 13 shows some results of interpolation of the 3D DVD. Figure 13a shows one isosurface for the geological dataset, and Fig. 13b shows respectively the interpolated volume, a slice from it and the isosurface (same value as the one from the DVD). The isosurface created from the DVD tends to follow the edges of the boundary voxels of the Voronoi cells, whereas in the interpolated volume the isosurface is smooth. This is because the values between the boundary voxels of two Voronoi cells can be relatively far apart, resulting in very distinct isosurfaces. In the interpolated volume however, the difference in values is spread over a larger area, resulting in smoother isosurfaces.

5 Conclusions We have shown that the 3D DVD provides a simple but effective tool for the modelling of 3D continuous datasets, and offers an alternative to using 3D grids or the 3D VD. It retains several of the advantages of the 3D VD, but has the advantages of

240


Fig. 13 (a) Isosurface created from the DVD. (b) The interpolated volume. Left: Its surface boundary. Centre: a slice from it. Right: an isosurface [same value as (a)]

being stored in a spatial model (grids) for which there are several tools available already. We have presented new algorithms to construct and dynamically modify 3D DVDs, these have the main advantage of being much less error-prone than the exact 3D VD and being conceptually simple (thus relatively simple to implement). The incremental construction, the dynamic deletion of seeds and the natural neighbour interpolation are all based on one operation that dilates spheres to update a DVD. It should however be said that, as presented in this paper, the use of the 3D DVD for point-based geoscientific datasets is somewhat limited. We however believe it to be the first step for the construction and the modelling of other types of VDs, such as the generalised VD or the VD for other primitives. Indeed, the VD discussed in this paper is the ordinary VD, which assigns volumes to points based on the Euclidean distance. There are however many more types of VDs that are also used in different fields. Okabe et al. (1994) show how, with the help of 12 different generalised VDs, 35 different neighbourhood operations can be performed. The VDs they propose are generalised in space, in the assignment function (which determines to which seed a location is assigned to), and in the set of seeds that are used. Examples of these are the weighted VD and the high order VD. These can be very difficult to construct or modify in vector format in 2D (Gahegan and Lee 2000; Dong 2008), let alone in 3D. It would be interesting to see if the discrete versions of these generalised VDs can also be created with the algorithm we proposed in this paper. Our first analysis tells us that minor changes to the algorithm would permit us to indeed create weighted VDs, but we have not implemented them yet. The 3D VD for other primitives such as lines, surfaces and polyhedra is also an interesting extension of this project since, as far as we know, no known (robust) algorithms currently exist.


241

References Anselin L (1999) Interactive techniques and exploratory spatial data analysis. In PA Longley, MF Goodchild, DJ Maguire, and DW Rhind, editors, Geographical Information Systems, second edition, pages 253–266. Wiley, New York. Bak PRG and Mill AJB (1989) Three dimensional representation in a Geoscientific Resource Management System for the minerals industry. In J Raper, editor, Three Dimensional Applications in Geographic Information Systems, pages 155–182. Taylor & Francis, London. Devillers O and Teillaud M (2003) Perturbations and vertex removal in a 3D Delaunay triangulation. In Proceedings of the 14th ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 313–319. Baltimore, MD. Dong P (2008) Generating and updating multiplicatively weighted Voronoi diagrams for point, line and polygon features in GIS. Computers and Geosciences, 34(4):411–421. Field DA (1986) Implementing Watson’s algorithm in three dimensions. In Proceedings of the 2nd Annual Symposium on Computational Geometry, pages 246–259. ACM Press, Yorktown Heights, New York. Fisher PF (1997) The pixel: A snare and a delusion. International Journal of Remote Sensing, 18 (3):679–685. Gahegan M and Lee I (2000) Data structures and algorithms to support interactive spatial analysis using dynamic Voronoi diagrams. Computers, Environment and Urban Systems, 24 (6):509–537. Gold CM (1989) Surface interpolation, spatial adjacency and GIS. In J Raper, editor, Three Dimensional Applications in Geographic Information Systems, pages 21–35. Taylor & Francis, London. Hoff KEI, Keyser J, Lin M, Manocha D, and Culver T (1999) Fast computation of generalized Voronoi diagrams using graphics hardware. In SIGGRAPH ’99: Proceedings of the 26th Annual Conference on Computer Graphics and Interactive Techniques, pages 277–286. ACM Press, New York. ISBN 0-201-485605. Hoffmann CM (1989) The problems of accuracy and robustness in geometric computation. IEEE Computer, 22:31–42. Jones CB (1989) Data structures for three-dimensional spatial information systems in geology. International Journal of Geographical Information Systems, 3(1):15–31. Kemp KK (1993) Environmental modeling with GIS: A strategy for dealing with spatial continuity. Technical Report 93-3, National Center for Geographic Information and Analysis, University of California, Santa Barbara. Ledoux H (2006) Modelling three-dimensional fields in geoscience with the Voronoi diagram and its dual. Ph.D. thesis, School of Computing, University of Glamorgan, Pontypridd, Wales, UK. Ledoux H and Gold CM (2008) Modelling three-dimensional geoscientific fields with the Voronoi diagram and its dual. International Journal of Geographical Information Science, 22 (5):547–574. Ledoux H, Gold CM, and Baciu G (2005) Flipping to robustly delete a vertex in a Delaunay tetrahedralization. In Proceedings of the International Conference on Computational Science and Its Applications – ICCSA 2005, volume 3480 of Lecture Notes in Computer Science, pages 737–747. Springer, Singapore. Li C, Chen J, and Li Z (1999) A raster-based method for computing Voronoi diagrams of spatial objects using dynamic distance transformation. International Journal of Geographical Information Science, 13(3):209–225. Mennis J, Viger R, and Tomlin CD (2005) Cubic map algebra functions for spatiotemporal analysis. Cartography and Geographic Information Science, 32(1):17–32. Okabe A, Boots B, and Sugihara K (1994) Nearest neighbourhood operations with generalized Voronoi diagrams: A review. International Journal of Geographical Information Systems, 8 (1):43–71.

242


Owen SJ (1992) An implementation of natural neighbor interpolation in three dimensions. Master’s thesis, Department of Civil Engineering, Brigham Young University, Provo, UT. Park SW, Linsen L, Kreylos O, Owens JD, and Hamann B (2006) Discrete Sibson interpolation. IEEE Transactions on Visualization and Computer Graphics, 12(2):243–253. Perez C and Traversoni L (1996) Finite element simulation of shallow waters using natural neighbors techniques. In Proceedings of the Computational Methods in Water Resources, pages 239–245. Southampton, UK. Rong G and Tan TS (2006) Jump flooding in GPU with applications to Voronoi diagram and distance transform. In Proceedings of the 2006 Symposium on Interactive 3D Graphics and Games, pages 109–116. ACM Press, New York. Rong G and Tan TS (2007) Variants of Jump Flooding Algorithm for Computing Discrete Voronoi Diagrams. In Proceedings of the 4th International Symposium on Voronoi Diagrams in Science and Engineering. IEEE Computer Society, Pontypridd, Wales, UK. Russel D (2007) Kinetic Data Structures in Practice. Ph.D. thesis, Department of Computer Science, Stanford University, Stanford. Sambridge M, Braun J, and McQueen H (1995) Geophysical parameterization and interpolation of irregular data using natural neighbours. Geophysical Journal International, 122:837–857. Schueller A (2007) A nearest neighbor sweep circle algorithm for computing discrete Voronoi tessellations. Journal of Mathematical Analysis and Applications, 336(2):1018–1025. Sibson R (1981) A brief description of natural neighbour interpolation. In V Barnett, editor, Interpreting Multivariate Data, pages 21–36. Wiley, New York. Sugihara K and Inagaki H (1995) Why is the 3D Delaunay triangulation difficult to construct? Information Processing Letters, 54:275–280. Tomlin CD (1983) A map algebra. In Proceedings of the 1983 Harvard Computer Graphics Conference, pages 127–150. Cambridge, MA. van der Putte T (2009) Using the discrete 3D Voronoi diagram for the modelling of 3D continuous information in geosciences. Master’s thesis, Geographical Information Management and Applications (GIMA), Delft University of Technology, Delft, The Netherlands. Watson DF (1992) Contouring: A guide to the analysis and display of spatial data. Pergamon Press, Oxford, UK. Zhao R, Li Z, Chen J, Gold CM, and Zhang Y (2002) A Hierarchical Raster Method for Computing Voronoi Diagrams Based on Quadtrees. In ICCS ’02: Proceedings of the International Conference on Computational Science, pages 1004–1013. Springer, London, UK.

Challenges in 3D Geo Information and Participatory Design and Decision Jan B.F. van Erp, Anita H.M. Cremers, and Judith M. Kessens

Abstract The scope of 3D geo information applications is broadening to include for instance crises management and training. These new applications are no longer restricted to expert users, but aimed at involving a multitude of (non-professional) stakeholders in a participatory design and decision process. This introduces new challenges for the user-system interaction that go beyond 3D visualization. Based on two use-scenarios in urban planning and crisis management, we identify important user-system research areas: intuitive data access and manipulation, multi-stakeholder participatory design and decision, and multisensory experience. We argue that future 3D geo information applications can benefit from the spin-in of Human Factors Engineering technology from domains such as medical visualization and serious gaming. We plead for starting a research community around user-system interaction for future 3D geo information applications that sets the research agenda and ensures that new applications benefit from advanced usersystem interaction technologies.

1 Introduction Last decade has seen vast advances in the technology involved in 3D geo information and GIS applications, for instance in data capture, analysis and management, in modeling and simulation, in 3D visualization and in interoperability. These advances open up possibilities for new applications to the degree that they can fundamentally change urban planning processes, crisis management and training tools. For instance, urban planning will not be limited to experts tackling specific and relatively static planning issues one by one. New applications are foreseen that allow multiple stakeholders to develop an integrated view on multidimensional planning issues and to participate in a dynamic design and decision process. It is our view that

J.B.F. van Erp (*), A.H.M. Cremers, and J.M. Kessens TNO Human Factors, Netherlands Organisation for Applied Scientific Research, Kampweg 5, 3769ZG Soesterberg, The Netherlands e-mail: [email protected], [email protected], [email protected]


243

244

J.B.F. van Erp et al.

this transition results in new challenges with respect to user-system interaction. These challenges include visualization of 2D/3D/4D data, visualization and interaction for laymen, facilitation of participatory design and decision, and issues such as multisensory experience, affective appraisal and perceived safety of virtual designs. In this paper, we approach these challenges on user-system interaction from a Human Factors Engineering perspective. We start by sketching two scenarios (use cases) describing how 3D geo information is already changing the urban planning and crisis management processes and will continue to do so in the future. We use these scenarios to identify important challenges and to illustrate what we can learn from the Human Factors Engineering domain and how advanced technology can contribute. We finish by presenting the reasons why Human Factors Engineering should get involved and by giving recommendations for near-term research. Herewith, we aim to start the process of creating a research community around user-system interaction for future 3D geo information applications.

2 Vision on the Role of 3D Geo Information in Participatory Design and Decision Processes: Two Scenarios We will sketch two scenarios that underlie our current and future research efforts in this domain. The first concerns urban planning, the second crisis management.

2.1 2.1.1

Urban Planning An Urban Planning Scenario

The urban planning co-design team consists of: Pierre (urban planner), Nancy (representing the inhabitants of the existing neighborhood), Ann (alderman), Mark (fire fighter commander) and Irene (facilitator). There is a plan to build a new block in an existing neighborhood, including a school building. All team members have their own ideas about the new block, which they express in their own terms. Pierre finds it important that the block has a unique identity, enhancing the image of the neighborhood. Nancy wants the neighborhood to be pleasant with a lot of empty space for recreation. Ann stresses the importance of stimulating vitality of the neighborhood. Mark is mainly concerned with safety issues. The participants meet for the first time in the design room. Irene invites them to play a game in which they get acquainted and find out about the ideas of the other team members. At first it is hard to grasp the terminology and to understand the meaning and impact of various ideas. Irene tries to enhance the shared understanding by means of storytelling. All participants go out to visit the neighborhood,

Challenges in 3D Geo Information and Participatory Design and Decision

245

in order to experience the current situation. They carry a mobile device that, based on the current location, offers comments about the location made by inhabitants of the neighborhood. This information adds to a shared understanding of the issues that play a role. After having returned to the design room, they all sit down around a large interactive table which displays the map of the neighborhood in 3D. They start to visualize the requirements of all parties. First they introduce a large block on the map, representing the new building block. Immediately, a number of built-in models, concerning safety, recreation space, traffic, noise, shade, etc. become active. Based on the location and dimensions of the block, they compute their outputs and display them in bar charts on top of the table, visible to everybody. All participants are allowed to adjust the priorities of various aspects. For example, the fire fighter commander recognizes that the street is too narrow to be safe, which requires a narrower building. This is adjusted immediately on the map. All participants understand the reason why this adjustment is necessary. The urban planner now decides to add some height to the block, which adds to the unique identity of the neighborhood. After this phase, Irene is satisfied about the increased shared understanding in the group. Now it is time to start detailing the design. The location of the school in the block is determined. Ann is happy with the vitality that the school will add to the neighborhood. However, Mark is very concerned about safety, since the school is too close to a busy road. He suggests moving it to the other side of the block. Nancy still finds it difficult to get a real feeling of what the location of the school will mean for the inhabitants. She goes out to experience the noise level at the planned school location (see Fig. 1), which is quiet now, but will be less quiet after the school and roads have been built. Walking around with a head phone playing the output of the noise calculation models, she finds the noise is way too loud. The other team members can follow her location on the map table and experience the same noise. They suggest turning the building in another direction and adding a row of

Fig. 1 Mixed reality technology lets stakeholders experience the effects of model output (e.g. noise) in the real world (left panel) or see the effects of planning alternatives in a virtual design (right panel)

246


trees, which has an immediate positive effect on the noise levels. Satisfied, Nancy returns to the design room. Now, the definite location and shape of the block, including the school, is determined. All participants can see the block in 3D. Also, all model outputs are visualized around the block. Participants can see that all requirements are met and they are satisfied with the result.

2.1.2

Current State-of-the-Art in Urban Planning

3D geo information is essential for the development of 3D city and urban planning tools. At TNO, we developed the spatial planning tool Urban Strategy (Borst et al. 2009a, b). In Urban planning, many topics from various domains play a role, such as mobility, noise, air quality, safety and aesthetics. Human health and comfort as well as climate change are topics that currently arise. Urban Strategy aims to integrate models of these different domains into a coherent picture. The integration of models from various domains poses a number of challenges: they need different input and output data, they have different levels of detail and abstraction, etc. For instance in Urban Strategy, the noise calculations require detailed and accurate information of the geography of roads, while the traffic model needs travel times from node to node on a coarser network. In order to view and manipulate the data, three interfaces were developed: 1D, 2D and 3D, as illustrated in Fig. 2. The 1D interface shows different indicators. The 2D interface is a GIS interface that enables the user to view, select and edit objects in the database. With this interface, the operator can define scenarios by adding, deleting or changing objects. Road attributes can be changed (such as the road surface type or speed of traffic), roads can be added or closed. The building configuration can be altered and noise barriers can be added. The 3D interface provides a 3D view of the city, including landmarks for orientation, and model output with a (3D) geographical component, such as noise contours. Urban Strategy has been successfully applied in various cases. Evaluation of these cases showed that urban planners were triggered to come up with concrete plans. The direct visualization of these resulted in new insights amongst the urban planners. Consequences of changes in a plan were explored in the same session. These results show that the use of Urban Strategy had the potential to change the

Fig. 2 Different visualizations used in urban strategy, from left to right: 1D indicators, 2D GIS interface, 3D overview display


247

planning process from a predominantly serial process to a more parallel, interactive and collaborative process. One important lesson that has been learned is that it is not the technology that forms a bottle-neck in time, but the Human Factors aspects. The authors conclude (Borst et al. 2009a): “It is a challenge to get all the different ideas that are generated during a session from the heads of the participants into the database. After ideas have been entered, the instrument provides output on a lot of different aspects. All this information is sometimes hard to grasp for each participant”.

2.1.3

The Future of Urban Planning: Urban Strategy 2.0

The goal of the scenario in Sect. 2.1.1 is to illustrate how we envision that co-design processes and the use of mixed reality will result in the next generation of urban planning tools. In the process of designing a building or planning an environment, many people, from different perspectives, should be able to provide input. This is called a “collaborative design” (co-design) approach. Our vision is that a mixed reality system is capable of facilitating that different people, from different perspectives in different phases of the co-design process, can communicate and collaborate to achieve a goal that is satisfactory for all participants. Mixed reality refers to the merging of real and virtual worlds to produce new environments and visualizations, where physical and digital objects co-exist and interact in real time. It may include a mix of reality, augmented reality (reality augmented with virtuality, such as model output), augmented virtuality (virtuality augmented with reality, such as photos) and virtuality.

2.2 2.2.1

Crisis Management A Crisis Management Scenario

It is a warm day during spring break. The emergency room receives an alarm about a fire at the camp site “Green Woods”. The environment of the camp site is a nature park with lots of woods and recreation facilities. The weather has been warm for quite some time now, so there is a severe risk of a forest fire. The situation at the camp site is chaotic; people are trying to find their way out, in the course of the process hindering each other as well as the first aid workers that have already arrived at the scene. People overrun each other and some casualties are reported. Soon there is a good piece of forest on fire. There appears to be a group of scouts on a weekend trip near the campsite. Their parents are worried about the safety of the children and start calling the authorities. If the wind direction changes slightly, a nearby elderly home will be in danger. As soon as the size of the incident becomes clear, a CoPI team (command at incident location) is formed consisting of Ann (alderman and deputy mayor), Mark

248


(fire fighter commander) and John (head of the regional emergency services). Together, they have to decide upon the best options to fight the crisis and minimize the number of victims. They stand around a touch table with a 3D depiction of the incident location and the current situation, including the locations of the fire, victims and available aid workers. This gives a quick glance on the current state of the crisis and facilitates building a shared understanding. Mark gives them a quick update and informs them on the expected spread of the fire. He does so by using the “fast-forward” option of the information system. Based on a multitude of data sources (such as the weather forecast, a database with the road capacities, local sensor data on temperatures and the chemical composition of the smoke) and models (such as fire and gas dispersion models), the fast forward button gives an impression of how the crisis will develop in the next 60 min. It turns out that the elderly home is indeed in danger. Now Mark, Ann and John start to develop alternative plans to fight the crisis by going through a number of if . . . then questions. Ann suggests concentrating all fire fighting capacity to protecting the elderly home. She does this by picking up the available fire trucks and moving them to the vicinity of the elderly home. Now, the fast-forward option shows that the camping grounds will be lost and there will be several victims in that area. Mark suggests evacuating the elderly home and concentrating the fire fighters near the campground. The evacuation model calculates that it will take about 35 min to clear the elderly home which is too long before the fire will be a threat to the elderly home itself and the route the evacuation vehicles need to take. By moving one fire fighter truck from the camping ground to the elderly home, the situation seems to be under control. Slowing the fire near the elderly home buys enough time to complete the evacuation safely while there is enough capacity available to control the fire at the camping ground and save the lives of the scouts. Ann, Mark and John decide that this is the best option and quickly start to give their orders.

2.2.2

Current State-of-the-Art in Crisis Management Tools

In case of such a disaster, firemen, policemen, medical aid workers, and local authorities are needed to fight it, each with their own focus and expertise. Together, they form a temporary multi-disciplinary crisis organization (the CoPI team). In a short period of time, it requires them to process a lot of information in order build up an image of the situation and to decide which actions should take place. The CoPI team operates at or near the location of the incident, and meets periodically in a mobile trailer which is specially equipped for this purpose. The team relies heavily on multidisciplinary information sharing and joint decision-making. Currently, CoPI teams have a plotter, who plots information on a 2D digital geographical map of the incident area and the information to be shared (see Fig. 3, left panel). This geographical plot is essential to support the creation of a shared image of the current situation between the disciplines. In addition, the plot supports the creation of future scenarios of the crisis (a rudimentary fast forward button)


249

Fig. 3 Crisis management is becoming an important 3D geo information application. Like urban planning, it is a multi-stakeholder issue that can be supported by intuitive interaction and predictive model visualization, especially when combined with real time sensor data depicting the current situation. The left panel shows the current 2D visualization in which information on victims etc. is displayed as a layer over a digital map. The right panel shows an advanced crisis management tool that allows tangible interaction and has options to predict outcomes of different alternatives

to help in taking precautionary measures. We highlight three innovative functionalities of the plot: thematic layers, current information and scenario layers.

Thematic Layers In order to stimulate multidisciplinary coordination, CoPI meetings are organized around various themes, such as fire development, security, victims, emergency services, public safety, and communication to the general public. Therefore, “thematic layers” are added to the plot, which support the organization of information according to these themes. Each layer, which can be accessed via a tab, contains only information that is relevant to the theme. This enables switching between various layers according to the themes that are being discussed, which protects participants from information overload.

Current Information A second innovation is that current, up-to-date information can be displayed in the plot via a wireless connection to intelligent devices and sensors in the field. For example, medical aid workers could be equipped with a handheld device for entering information about victims (location, identity, triage). Subsequently, victims could be provided with an RFID tag to monitor their locations in case of displacement. Moreover, victims could be equipped with a special “patch”, which monitors vital bodily functions (heart rate, blood pressure, temperature). In the CoPI trailer, the victims’ current locations could be visualized by icons on the plot, giving access to their personal details and current physical state. The system could automatically keep

250


track of the total number of victims encountered so far. Also, locations of fire trucks, police cars and ambulances can be determined through a GPS system and made visible on the plot. A third example is sensors for toxic fumes, which could be attached to cars and trucks or to uniforms of aid workers. By collecting this information from sensors that are spread over the field, an accurate image of the toxic cloud could be drawn, and precautionary measures could be taken fast.

Scenario Layers The wind direction and speed are of great influence on the development of a forest fire. Local meteorological predictive information can be collected automatically. Expansion of the fire can be calculated through available models, tailored to the specific circumstances of the area. In this manner, various scenarios can be calculated and visualized on the plot for different moments in the future. This will provide the CoPI team with tools to decide on precautionary measures. Alternative scenarios can be plotted in different plot layers.

2.2.3

The Future of Crisis Management

Using a similar platform as described in the urban planning scenario, we plan to expand the current predominantly 2D crisis management tool to a 4D tangible geo information based support tool (see Fig. 3, right panel). Especially in urbanized areas, full integration of geospatial information is of eminent importance, including information on buildings and how many people may be present at what time of day, the capacity of roads and the fastest routes for aid workers, areas that are prone to flooding, etc. Also, we are building a software architecture that allows us to couple heterogeneous, predictive models across domains. This means that the effects of different options are calculated for all relevant parameters and we can realize real 4D visualization. As with the urban planning platform, we aim to make this tool easily accessible for people who are not trained on the system, for instance by allowing tangible interaction like physically placing fire trucks on the map or blocking a road.

3 User-System Interaction Challenges A way of taking into account Human Factors Engineering aspects during the design of 3D geo applications is to apply a usability engineering method (Gould and Lewis 1985; Nielsen 1994). Usability engineering is based on three principles: (1) early and continuous focus on user and tasks, (2) empirical measurement, (3) iterative design. Our Usability Engineering (UE) method (Neerincx and Lindenberg 2005) has been developed and applied for space missions, ship control centers, and mobile services. The method starts with the definition of a concept, which is a broad


251

description of the proposed system. Scenarios are then drafted from the relevant application domain and describe users, their tasks and context in a comprehensive, narrative style. From the scenarios, the process of requirements analysis results in a requirements specification. These requirements describe in detail the user needs with respect to their work practice and the role the system fulfils in addressing these needs. User requirements form the basis for the system features. Features can be considered solutions to user needs and describe what functionality the system should have. Evaluation is done by through objective and subjective quality criteria, such as effectiveness, efficiency, satisfaction and domain-specific performance criteria. We discuss three challenges in the field of Human Factors Engineering: intuitive interaction, co-design and decision, and multisensory experience. Most challenges apply to both urban planning and crisis management, examples are taken from the former.

3.1

Intuitive Data Access and Manipulation

Broadening the scope of 3D geo information applications means that not only expert users need access to the system but also non-professionals like politicians and safety officers, and even the general population (Counsell et al. 2009; Lai et al. 2010). This implies that people without domain knowledge or even basic computer skills should be able to assess, understand and manipulate the information. This requirement is relevant in many Human–Computer interaction applications and also recognized in urban planning applications. Generally, we see a trend towards intuitive interaction, meaning that the interface can be used without specific training or computer skills (Lu et al. 2009). A typical example is the Nintendo Wii controller, showing that replacing complicated joystick interfaces with touch or movement-based interfaces can provide access to complicated systems such as game computers for a much wider range of users than ever before. Adding or changing objects on the map (and consequently in the linked models and databases) can be achieved in either a virtual or a physical way (Ishii 2008). In the virtual approach, this is done through a graphical user interface (GUI). GUIs represent information in the form of pixels on displays. These graphical representations are manipulated with generic remote controllers (such as mice and keyboards). By decoupling representation (pixels) from control (input devices) this way, GUIs are malleable enough to graphically emulate a variety of media. However, when interacting with the GUI world, we cannot take advantage of our evolved dexterity or utilize our skills in manipulating physical objects (such as building blocks or clay models). Furthermore, coupling of the movement (direction) of the control device and of that of objects on the GUI may be confusing, especially when a third dimension is added to a 2D control (mouse) and display (projection screen) (Van Erp and Oving 2002). In the physical approach, tangible user interfaces can be used. Physical objects, such as building blocks can be placed on

252


the map. The system recognizes the object as well as its features, which can be used to make computations (e.g. to apply models). By a giving a tangible (physical) representation to digital information, tangible user interfaces make information directly graspable and controllable through haptic feedback (Van Erp 2002). They also provide a strong 3D effect. Tangible user interfaces are considered to be more intuitive for laymen. Intangible representations (such as video projection) may complement the tangible representation, synchronizing with it. We see a similar trend towards intuitive interfaces with regard to advanced 3D and 4D visualization technologies. These are less and less based on desktop displays (e.g., Jones et al. 2009), therefore requiring careful consideration of control device-display mapping (Van Erp and Oving 2002) or even new ways to interact and manipulate objects, other than mouse and keyboard (Ibrahim and Noor 2004). Human Factors Engineering expertise can provide the relevant guidelines for these developments. Both technological and user issues play a role, for instance: l

l

l

l

l

l

Intuitive panning (moving the point of view), zooming (changing the level of detail), and information display. For instance, Roscoe (1968) gives guidelines on display integration (that state that it is beneficial to integrate indicators into a single presentation) and pictorial realism [that state that graphically encoded information (symbols) can be readily identified with what they represent]. Clutter reduction. Clutter refers to the state in which excess items, or their representations or organization, lead to a degradation of performance at some task (Baudisch and Rosenholtz 2003). To avoid clutter, one can change the level of detail (e.g. nested level of detail, or location dependent level of detail), use enhancement effects (e.g. pop-out effects), aggregate objects, or remove irrelevant information. 2D vs. 3D presentation of (map) data. Some usability tests show that 3D maps improve task performance compared to 2D maps, because spatial relationships are better understood (Rakkolainen and Vainio 2001; Warren and Wertheim 1990). However, when a 3D view is displayed on a 2D screen, one or more dimensions are compressed depending on the elevation angle (e.g., an elevation angle of 90 (looking down) results in a 2D display; Van Erp and Kappe´ 1997). A general disadvantage of 3D representation is that the overview can be lost easier and details are less clearly visible compared to 2D representation (Looije et al. 2007). Map orientation. Depending on the task or individual user, maps should be oriented north-up (i.e. world referenced) or forward-up (i.e. ego centered) (Darken and Cevik 1999; Van Erp and Kappe´ 1997). Location dependent information presentation. By using sensors, information on the user location or environment can be automatically derived and the application can be adapted accordingly. Multi-task environments. As the user can perform different tasks with 3D geo information applications while he/she is also performing other tasks, the user interface should be adapted accordingly. Examples include car navigation systems that display route information not only visually but also by a synthetic voice or even vibration (Van Erp and Van Veen 2004).

Challenges in 3D Geo Information and Participatory Design and Decision l

l

253

Personalization. An application should preferably be personalized to the specific user. To make the right decisions on how the application should adapt, it should have knowledge about the user such as: preferences, skills, and knowledge. This is usually done by using user models, e.g. standard user models (Fink and Kobsa 2002) and models based on implicit or explicit acquisition of information (Cremers et al. 2002; Van Erp et al. 2010). Advanced visualization. We should realize a spin-in of relevant technology on the use of stereoscopic displays, Virtual Reality (see also below) and other advanced visualization technologies. The same holds for the manipulation side of the interaction. Recently, (multi-) touch tables making the interaction more tangible are also being explored for 3D geo information applications (e.g., Asai et al. 2008). Maher et al. (2005) showed that the use of tangible user interfaces increases shared understanding and leads to more creativity compared to graphical user interfaces.

3.2

Multi-stakeholder Participatory Design Processes

The various co-design team participants, including end-users such as civilians, have different backgrounds, interests and goals. At the start of the process, the understanding of the problem can be more diverse, at the end it should be shared. A facilitator should be added to the team, to monitor whether shared understanding and satisfaction is achieved in the course of the process, and to take action if this not the case. Also, the facilitator should keep track of decisions made, including underlying arguments, in order to support efficiency of the process. The system should allow multisensory “visualization” of the design in different stages. Participants should be able to experience the current situation, the different perspectives of the participants on the situation, and the various solutions in different stages of the design. It is essential to make the experience as realistic as possible, since it is not possible to experience future design alternatives in real life. Experiencing design alternatives together helps to evaluate them, by determining how specific aspects of each solution meet the various interests and experiences of the participants. Mixed reality can be applied in different ways during the process: l

l

l

It can be used to bring the real world to the co-design meeting, by displaying photos and webcam images on a 3D interactive map or playing sounds of the current situation. It can bring the virtual world to the meeting, by visualizing output of models and simulations on the 3D map in the meeting room. It can bring the meeting to the real world, by augmenting current reality with virtual elements, such as various future design alternatives and model outputs.

A main challenge in the future of urban planning is to make the planning process simpler, faster and more efficient. We follow a twofold approach here. The first is to develop a tool that enables early integration over stakeholders and experts to

254


improve the current way of working based on late integration (see Sect. 3.2.1). The second is to compress the time line by combining phases in the planning process (see Sect. 3.2.2).

3.2.1

Compress Over Disciplines: Involving Stakeholders and Experts

Increasingly relevant is the awareness that urban planning is a multi-stakeholder process (Aspin and Roberts 2005) with involvement of politicians, architects, builders, environmentalists, civilians, etc. (Isaacs et al. 2008). Efficient planning depends on involvement of these stakeholders as early as possible (see Fig. 4). This has several consequences. First, these stakeholders are mostly non-experts in urban planning (tools) and may experience difficulties in getting truly involved in the design and decision process and in understanding and experiencing the planning results (Hetherington et al. 2007). This may reduce their support for the process and the outcome. Second, successful applications go beyond supporting visualization and data processing, but also support the process. Some important factors or phases in a successful co-design and decision process are: getting acquainted, building mutual trust and balancing the contribution of participants (e.g. speaking time). A proficient human facilitator will monitor these factors, but there are also relevant technological tools that can (further) enhance the process. Examples include: creating an atmosphere that matches the process phase, for instance through manipulation of room color and odor, using simple gaming technologies to build trust among the participants or expressing ideas or desires, and automatically

Fig. 4 The goal of participatory design is to include all stakeholders as early as possible to change the current serial process (left panel) into a parallel process (right panel)


255

monitor the group process through registration of aspects such as turn taking, interruptions, etc. and feed this information back to the participants (Cremers et al. 2007). We are currently working on AI and interface technology needed to implement a virtual coach that facilitates and supports the co-design and decision process which we call the Virtual Facilitator. 3.2.2

Compress Over Time: Fast Forward Scenarios and Real-Time Sensor Data

An interesting new development is to use predictive models in forecasting developments as function of different options (Wang et al. 2008) across domains. This is of utmost importance in crisis management but also relevant for urban planning. Integration fast forward models over domains such as environment, economics and crime (Peng and Jones 2004), results in a planning tool that can predict the effects of design and decision alternatives (near) real time. This has the potential to radically change the planning process from a predominantly serial process to a parallel process with fast iterations of concept development, scenario generation, and planning stages (Yao et al. 2006). In combination with participatory design, this may turn urban planning from a sequential process into a parallel process in which all stakeholders are interactively involved from the beginning. An important prerequisite is that the fast-forward effects of scenarios are calculated relatively fast, i.e. in the timeframe of a workshop. A related development is the inclusion of sensor data to assess the real-time situation and combine this real world data with data from the virtual world (e.g., design choices, model predictions). This form of mixed reality is especially relevant for crisis management applications, which is also a multi-stakeholder issue. Mixed reality may open up new possibilities of intuitive visualization and interaction, but also challenges the user to integrate abstract (semantic) data, real world images and the virtual representation of objects into one coherent representation. A challenge that requires new visualization solutions such as Sedlmair et al. (2009) proposed. They added an additional virtual representation of the (real world) object to illustrate the connection between abstract and real-world issues. A similar approach was taken by Einsfeld et al. (2008) who designed a modified Virtual Reality to integrate semantic information directly into the scene facilitating non-expert users.

3.3

Multisensory Experience

As already indicated by Hetherington et al. (2007), providing the general public and the planning authority with a good understanding of the visual impact of proposals is important. In their study, they restricted the impact to the visual modality through the creation and (web) presentation of visual models of proposals that could be viewed from different angles. However, urban planning proposals have a much wider impact than only visual, including sound (noise), smell, and vibration, but

256


Fig. 5 The goal of a multisensory mixed reality is to draw participants really into the design to let them experience the design alternatives as they would eventually be experienced in the real world

also on for example perceived safety. Some of the models to calculate these aspects are already available and some are still under development. One can however imagine that assessing the impact of a new highway and/or noise reduction measures based on simulating and experiencing the actual noise has a much larger impact than presenting abstract values or noise contours. Apart from the challenge of building accurate models, there is a big challenge in developing the technology that is able to let the user experience the correct multisensory impact of proposals. This certainly holds for other than the visual modality, but also experiencing the visual modality can be improved. For instance by using Augmented Reality technology such that stakeholders can experience the plans in their real environment and context and from the correct perspective (see Fig. 5). Also, artist impression like drawings and models will not necessarily result in the same experience as the actual situation. It is not always a sunny day, not all trees look the same (or are even there) and not all walls are without graffiti (Toet et al. 2009).

4 Conclusion and Recommendations One of the conclusions from the 2006 workshop on 3D geo information was that about 90% of the problems in 3D data visualization were solved (Rahman 2006). One might have expected that 4 years would have been enough to solve the final


257

10%. However, increasing availability of 3D geo information also resulted in involvement of domains other than landscaping and urban planning, for instance safety and crisis management. One of the prerequisites for the successful introduction of these new application domains of 3D geo information is advanced usersystem interaction that goes beyond that of 3D or 4D data visualization. Our vision is that a mixed reality system is capable of facilitating that different stakeholders, with different perspectives in different phases of the co-design process, can communicate and collaborate to achieve a goal that is satisfactory for all participants. We argue that the field of 3D geo information can benefit from Human Factors Engineering technologies from other domains such as multimodal interaction and gaming technology, because: l

l

l

There is limited spin-in from visualization technology in for instance medical visualization and command-and-control applications. The 3D geo information community could benefit from the relevant knowledge and guidelines established in these domains. Broadening the 3D geo information application range turns the user group into a multidisciplinary one and makes one-size-fits-all solutions unsuited. Also, users will not all be experts on the system and/or have the possibilities to follow a training in system handling skills. Human Factors Engineering is used to developing solutions for a diversity of users or user groups and in the design of selfexplaining and intuitive systems. New 3D geo information applications may cause a revolution in the way design and decision processes are organized. Human Factors Engineering includes the development of technology to facilitate these processes.

Therefore, we aim to organize a research community around user-system interaction for future 3D geo information applications. Focus points of coordinated research may be the development of intuitive data access (including various ways of data visualization 1D/2D/3D/4D and their effects on shared understanding and satisfaction of the participants), data manipulation (e.g. the use of more intuitive interaction like a MultiTouch Table), facilitating multi-stakeholder participatory design and decision to combine domains and compress process duration, and multisensory experience as argued in the main body of this paper.

References Asai K, Kondo T, Kobayashi H, Mizuki A (2008) A geographic surface browsing tool using mapbased augmented reality. Proceedings International Conference Visualisation, VIS 2008, Visualisation in Built and Rural Environments, pp 93–98, London, United Kingdom, 9–11 July 2008 Aspin R, Roberts D (2005) A model for distributed, co-located interaction in urban design/review visualisation. Proceedings IEEE International Symposium on Distributed Simulation and RealTime Applications, DS-RT, pp 70–77, Montreal, Canada, 10–12 Oct. 2005 Baudisch P, Rosenholtz R (2003) Halo: a technique for visualizing off-screen locations. Proceedings of CHI 2003, pp 481–488, Ft. Lauderdale, Florida, USA, April 5–10 2003

258


Borst HC, Schelling AL, Lohman WLA, Minderhout M, Salomons E, Meijer E, Boot H, Van Gijlswijk R, Klerkx R (2009a) Urban strategy: instrument for interactive sustainable planning. 15th Annual International Sustainable Development Research Conference, Utrecht, The Netherlands, July 5–8, 2009 Borst HC, Salomons EM, Lohman WJA, Zhou H, Miedema HME (2009b) Urban strategy: noise mapping in instrument for interactive spatial planning. Proceedings of Euronoise 2009, Edinburgh, Scotland, 26–27 October 2009 Counsell J, Richman A, Holding A (2009) Evaluation of 3D visualisation in the virtual environmental planning systems project. Proceedings of the 2nd International Conference in Visualisation, VIZ 2009, pp 108–113, Barcelona, Spain, 15–17 July 2009 Cremers AHM, Groenewegen P, Kuiper I, Post W (2007) The project browser: supporting information access for a project team. HCII July 25–27 2007, Beijing, China Cremers AHM, Lindenberg J, Neerincx MA (2002) Apples or oranges: a user-centred framework for cooperative user profile management. 7th WWRF Meeting, Eindhoven, The Netherlands, 3–4 December 2002 Darken RP, Cevik H (1999) Map usage in virtual environments: orientation issues. Proceedings of IEEE VR, pp 133–140 Einsfeld K, Ebert A, W€ olle J (2008) Modified virtual reality for intuitive semantic information visualization. Proceedings of the International Conference on Information Visualisation, pp 515–520, London, UK, 9–11 July 2008 Fink J, Kobsa A (2002) User modeling for personalized city tours. Artificial Intelligence Review 18: 33–74 Gould JD, Lewis C (1985) Designing for usability: key principles and what designers think. Communications of the ACM 28: 300–311 Hetherington R, Farrimond B, Clynch P (2007) Interactive web visualisation of proposals for site developments. Proceedings of the International Conference on Information Visualisation, pp 613–619, Zurich, Switzerland, 4–6 July 2007 Ibrahim N, Noor NFM (2004) Navigation technique in 3D information visualisation. IEEE Region 10 Annual International Conference, Proceedings/TENCON, B, pp B379–B382, Chiang Mai, Thailand, 21–24 Nov. 2004 Isaacs J, Falconer R, Blackwood DA (2008) Unique approach to visualising sustainability in the built environment. Proceedings International Conference Visualisation, VIS 2008, Visualisation in Built and Rural Environments, pp 3–10, London, UK, 9–11 July 2008 Ishii H (2008) The tangible user interface and its evolution. Communications of the ACM 51(6): 32–36 Jones RR, McCaffrey KJW, Clegg P, Wilson RW, Holliman NS, Holdsworth RE, Imber J, Waggott S (2009) Integration of regional to outcrop digital data: 3D visualisation of multiscale geological models. Computers and Geosciences 35(1): 4–18 Lai PC, Kwong KH, Mak ASH (2010) Assessing the applicability and effectiveness of 3D visualisation in environmental impact assessment. Environment and Planning B: Planning and Design 37(2): 221–233 Looije R, Te Brake GM, Neerincx MA (2007) Usability of mobile devices and services table of contents. Proceedings of the 4th International Conference on Mobile Technology, Applications, and Systems and the 1st International Symposium on Computer Human Interaction in Mobile Technology, ACM, New York, pp 532–539, Singapore, September 12–14, 2007 Lu G, Shark LK, Hall G, Zeshan U (2009) Dynamic hand gesture tracking and recognition for realtime immersive virtual object manipulation. International Conference on CyberWorlds, CW’09, pp 29–35, Bradford, West Yorkshire, UK, 7–11 Sept. 2009 Maher ML, Kim MJ (2005) Do Tangible User Interfaces Impact Spatial Cognition in Collaborative Design? Cooperative Design, Visualisation, and Engineering, LNCS, pp 30–41 Neerincx MA, Lindenberg J (2005) Situated cognitive engineering for complex task environments. 7th International Conference on Naturalistic Decision Making. Amsterdam, The Netherlands Nielsen J (1994) Usability Engineering. San Francisco, CA: Morgan Kaufmann


259

Peng C, Blundell Jones P (2004) Reconstructing urban contexts online for interactive urban designs. Design Studies 25(2): 175–192 Rahman AA (2006) Workshop 3D geo-information 2006. GIM International 20(10): 49 Rakkolainen I, Vainio T (2001) A 3D City Info for mobile users. Computers and Graphics 25: 619–625 Roscoe SN (1968) Airborne displays for flight and navigation. Human Factors 10(4): 321–332 Sedlmair M, Ruhland K, Hennecke F, Butz A, Bioletti S, O’Sullivan C (2009) Towards the big picture: enriching 3d models with information visualisation and vice versa. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 5531 LNCS, pp 27–39 Toet A, Van Welie M, Houtkamp J (2009) Is a dark virtual environment scary? Cyberpsychology and Behavior 12(4): 363–371 Van Erp JBF (2002) Guidelines for the use of vibro-tactile displays in human computer interaction. Proceedings of Eurohaptics 2002, pp 18–22 Van Erp JBF, Kappe´ B (1997) Ecological display design for the control of unmanned airframes. Proceedings of HCI International, Elsevier, Amsterdam, pp 267–270, San Francisco, USA, August 24–29, 1997 Van Erp JBF, Oving AB (2002) Control performance with three translational degrees of freedom. Human Factors 44(1): 144–155 Van Erp JBF, Van Veen HAHC (2004) Vibrotactile in-vehicle navigation system. Transportation Research Part F: Traffic Psychology and Behaviour 7(4–5): 247–256 Van Erp JBF, Veltman JA Grootjen M (2010) Brain-based indices for user system symbiosis. In: DS Tan, A Nijholt (Eds.) Brain-Computer Interfaces, Human–Computer Interaction Series, pp. 201–219. London: Spinger-Verlag Wang X, Shi J, Huang H (2008) Traffic-related air quality analysis and visualisation. Civil Engineering and Environmental Systems 25(2): 157–166 Warren R and Wertheim AH (1990) Perception and Control of Self-Motion. Hillsdale, NJ: Lawrence Erlbaum Associates Yao J, Tawfik H, Fernando T (2006) A GIS based virtual urban simulation environment. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 3993 LNCS III, pp 60–68

.

An Integrated Framework for Reconstructing Full 3D Building Models Langyue Wang and Gunho Sohn

Abstract Nowadays, in contrast to the increasing needs for 3D building models reconstructed with both indoor and outdoor structure and semantic information, the current state-of-the-art methods are not able to reconstruct the indoor and outdoor of buildings as a whole in one workflow. This paper proposes an integrated framework for the reconstruction of full 3D building models, in which both 3D indoor and outdoor model are reconstructed in a collaborative manner by fusing airborne laser scanning (ALS) data, terrestrial laser scanning (TLS) data and architectural plans. As a proof of concept, a preliminary test to reconstruct integrated 3D facade and indoor model is presented. First, based on plane sweep technique, a semantic and geometric information integrated point matching based method is developed to register 2D floor plans with TLS points. Then based on the registration, 3D façade model and indoor model are reconstructed and integrated simultaneously. The test results demonstrate the feasibility of the proposed framework.

1 Introduction As the predominant object comprising cityscape, 3D building models are increasingly requested for a larger number of applications such as various urban and environmental application, tourism, public security, etc. Especially, with the rapid development of systems and applications for 3D surveillance and navigation, there is an urgent demand to develop automatic techniques for integrating 3D outdoor building models with indoor information. The challenge is to produce semantically, geometrically, and topologically correct 3D building models with both indoor and outdoor structure and information. Nowadays, a number of approaches have been proposed to automatically or semiautomatically reconstruct 3D building models based on various input data such as airborne images, close-range images, ALS data, TLS data, etc. Despite the promising

L. Wang (*) and G. Sohn GeoICT Laboratory, York University, Toronto, ON, Canada M3J 1P3 e-mail: [email protected], [email protected]


261

262

L. Wang and G. Sohn

results achieved by many researchers, one common limitation can be seen is that most approaches only focus on the reconstruction of parts of the building separately, for example rooftop (Suveg and Vosselman 2004; Brenner 2000), façade (M€uller et al. 2007; Becker 2009) and indoor (Lu et al. 2005; Or et al. 2005). To our knowledge, there is no technique available which produces a complete model of rooftop, façade and indoor structures attributed with semantic information within a single approach. A primary concern caused by this is that considerable extra efforts have to be taken to integrate all the partial building models in order to make a complete building representation. In most cases, the different accuracies and data structures of the partial models could make the integration very difficult. Another problem faced by existing methods is that the reconstructed 3D building models lack rich semantic information and topological relationships; whereas the success of most of the 3D building model based applications, especially for those based on indoor models, highly relies on the semantic information of and the topological relations between the 3D building components. As a consequence of these problems, despite the fact that a large number of 3D building models have been reconstructed, applications that can really take advantage of those building models are very few. For example, currently most of the emergency evacuation and navigation programs still mainly use 2D plans for visualization and communication and interiors of buildings are often represented as two-dimensional spaces with attributes attached to them (Meijers et al. 2005). As can be seen, it will be of great benefit if 3D building models with both indoor and outdoor semantic and geometric information can be reconstructed in one framework. In order to differentiate this kind of building models from the above-mentioned models, we call these building models full 3D building models. Based on the stateof-the-art 3D building reconstruction techniques, this paper proposes an integrated framework for full 3D building model reconstruction by fusing ALS data, TLS data and 2D architectural plans. In next section, as the inspiration to the proposed framework, the related upto-date research works are briefly introduced. The existing problems and the opportunities created by these researches are analyzed, based on which the framework is presented in Sect. 3. Section 4 presents a method for registering 2D floor plans with TLS data and an implementation of the integrated 3D façade and indoor model reconstruction by fusing TLS data and 2D architectural floor plans. Preliminary result of an integrated 3D facade and indoor model is provided, which can be served as a demonstration of the feasibility of the proposed framework.

2 Related Works: Problems and Opportunities With the rapid development of laser scanning technology, using airborne and terrestrial laser scanning data has shown great potential towards automated 3D roof and façade reconstruction, especially when additional information, like 2D map data, is integrated. For 3D indoor modeling, using 2D floor plans has proven to be a practical way compared to other methods.

An Integrated Framework for Reconstructing Full 3D Building Models

2.1

263

Using TLS Data

The advantage of TLS data comes from its high geometric accuracy. But TLS data alone, it is usually not a trivial problem to extract correct semantic information and meaningful structures. Therefore, the main task of façade reconstruction using TLS data is to recognize the façade features and delineate their boundaries accurately. In Ripperda and Brenner (2009), a façade shape grammar is introduced to integrate the structure of façades in the stochastic reconstruction process. The façade elements are searched using a reversible jump Markov Chain Monte Carlo (rjMCMC) approach. An MDL based scoring function is used to score the derivation trees. This method is particularly suitable for regular shaped building facades. In Pu (2010), a knowledge based reconstruction of building models (mainly façade models) based on TLS data is presented, in which knowledge about the sizes, positions, orientations and topology is introduced to recognize features like walls, windows, doors, etc. Despite the promising results shown by the TLS data based methods, limitations of using TLS data include: l

l

l

l

Occluded features cannot be reconstructed reliably. Most of the methods cannot produce good results for uncertain and missing data areas caused by occlusion. Most of the methods can only detect façade features with regular shapes on planar walls and will fail in detecting complex building structures, such as windows with arc edges. To accurately delineate the boundaries of the façade features like windows, additional data sources are needed. For high buildings, sometimes the facades can hardly be covered completely by TLS points.

It is worth noting that using oblique airborne laser scanning data may overcome some of the problems, and therefore it could be a valuable resource for both facade and rooftop model reconstruction.

2.2

Using 2D Floor Plans

One big advantage to use 2D floor plans is that detailed indoor structures with semantic information can be reconstructed, even for complex structures. In Or et al. (2005), the authors present an efficient approach to quickly generate 3D geometric indoor models from floor plan images. In Horna et al. (2009), an operational way is presented to build 3D indoor topology from 2D architectural plans by using a formal description of constraints which provides a generic representation of geometry, topology and semantics of architectural indoor environments. Using these constraints, 3D indoor models can also be reconstructed automatically from 2D vector plans.

264

L. Wang and G. Sohn

But until now, only a few systems can fully address the problem of reconstructing 3D building models from 2D architectural plans although not completely automated (Yin et al. 2009). One common problem that all the methods using 2D floor plans for generation of 3D building models face is the lack of geographical coordinates. Furthermore, individual floor plans may have different origins, scales and orientations. Because in most of the methods 3D indoor models are reconstructed by extruding the 2D walls vertically, buildings with non-vertical walls can hardly be extruded without other clues.

2.3

Using ALS Data and 2D Map Data

In Kada and McKinley (2009), the authors present a reconstruction method to produce LOD2 models (roof models) from existing ground plan and ALS data based on a 2D partitioning algorithm that splits a building’s footprint into quadrangular sections. Then models are constructed by assembling building blocks from a library of parameterized standard shapes. The method has been successfully used for many commercial projects. In Oude Elberink (2010), target graph matching algorithm is used to reconstruct 3D roof models. Targets are topological representations of the most common roof structures which are stored in a database. Existing 2D map data with scale 1:1,000 has been used for selection of building segments, for outlining flat building roofs and to reconstruct walls. Both methods show a highly automated reconstruction and can be used for large areas. One obvious limitation is that roofs with complex or irregular shaped structures are hard to be reconstructed correctly due to the lack of more useful information. And when looking at a higher level of detail, it becomes more complex. Form the promising results by using ALS data and 2D map data, it can be inferred that it will be more beneficial if 2D architectural plans can be used in the process of 3D roof model reconstruction since 2D architectural plans are more accurate and have more detailed information than 2D map data. Furthermore, although using TLS data can automate the process of 3D façade reconstruction, the lack of geometric and semantic information still causes many problems. Therefore, the integration of TLS data and 2D architectural plans, although challenging, could resolve many practical problems in 3D building façade and indoor modeling and create an opportunity to reconstruct integrated 3D indoor and façade models with high geographic accuracy and rich semantic information. It can be reasoned out that the fusion of ALS data, TLS data and 2D architectural plans can create a way to reconstruct full 3D building models with roof, façade and indoor structure and semantic information within one framework.

An Integrated Framework for Reconstructing Full 3D Building Models Terrestrial laser scanning data

265

Airborne laser scanning data

Architectural plans

Registration

Georeferenced Architectural plans

Integrated 3D façade and indoor reconstruction

Integrated 3D roof reconstruction

Model integration

3D building models

Fig. 1 Proposed framework for full 3D building model reconstruction

3 The Framework for Integrated Indoor and Outdoor Reconstruction Figure 1 gives an overview of our proposed framework for full 3D building model reconstruction. To make it general, we group and categorize all the processing steps into four main processes.

3.1

Input Data

The input airborne and TLS data are georeferenced point clouds in the same geographic coordinate system. There are many kinds of architectural drawings in the architecture engineering and construction (AEC) industry like longitudinal-section drawings, elevation drawings, and reflective ceiling plans. Among them, most drawings take the form of floor plans, which are orthographically top-down projected portrays of each building level using standardized symbolic representations of the structure’s architectural elements (Yin et al. 2009). Architectural plans can be in paper format and digital format; and paper drawings can be digitized and converted to digital

266

L. Wang and G. Sohn

CAD files. It is obvious that using more types of architectural drawings will provide more useful information for the reconstruction, but the workload will also be increased. For a building, sometimes it may have no other architectural drawings except for floor plans. In this respect, methods using digital 2D CAD-based architectural floor plans could be more general. We will use floor plans for demonstrating the frame work. When combining other types of architectural drawings, some related processing will vary accordingly but the framework will remain a common framework.

3.2 3.2.1

Main Processes Registration of TLS Data and Architectural Plans

Because almost all architectural plans are not in the geographic coordinate system, the main objective of the registration is to georeference the architectural plans and aligns the two datasets in the same coordinate system so that further processes can be performed. After registration, all floor plans will be vertically arranged with the same direction and scale with accurate horizontal coordinates. Problems of the registration lie in the dimensional difference and the uncertainties or ambiguities intrinsic to the two datasets. Algorithms for the registration should be able to reduce or eliminate the ambiguity caused by occlusion and find and use all the potential reliable corresponding features for the calculation of the coordinate transformation. A good way is to using a sweep plane to scan the TLS data to find corresponding information between the two datasets. In Sect. 4.1, a sweep plane based registration method will be demonstrated. 3.2.2

Integrated 3D Façade and Indoor Reconstruction

After registration, architectural floor plans are aligned with the TLS points. The horizontal extents and locations of the façade features like windows and doors become known. Therefore, all the façade features that cannot be detected from TLS points due to occlusion can be located and at the same time, if required, all the points can be classified and no-data areas can be filled up according to the semantic information from floor plans. Then, based on the correct horizontal locations of the façade features, sweep line algorithm can be employed to detect the vertical extents of the façade features. The intrusions of windows and doors can be obtained either from floor plans or by estimating the laser scanning points inside the extents of the windows or doors. Thus the façade features can be reconstructed and located accurately. The storey heights can be obtained from either the architectural plans or by calculating the vertical distance between two windows of neighboring floors. Then the floor plans can be extruded to reconstruct the 3D indoor models. Protrusions, intrusions and other structures attached to the façade can also be reconstructed by combing the structure detected from TLS points and information from floor plans, if any exist.


3.2.3

267

Integrated 3D Roof Reconstruction

Existing approaches using ALS points and 2D map data provide much useful knowledge for using 2D architectural plans. Compared to 2D map data, in which buildings have only outlines, floor plans can provide more accurate and detailed information, especially the detailed partitions of the buildings. Based on these advantages, the existing methods can be much improved and new methods can also be developed if more architectural information is used. 3.2.4

Model Integration

The goal of this process is to integrate the roof model with the façade and indoor model into a full 3D building model. The accuracy difference between airborne and TLS data could cause some problems in the integration. Because the roof model and the integrated façade and indoor model are reconstructed based on the same registered floor plans, there should be no problem in terms of horizontal integration. The roof outline derived from floor plans can be used as a bridge to register the two datasets if the horizontal or vertical difference is bigger than the required threshold. Another problem is the geometric, semantic, and/or topological fusion between the roof model with the integrated façade and indoor model, algorithms may need to take the data structures of the two models into account.

4 Result The resulting 3D building model will be a complete building model with roof, façade and indoor structure and semantic information. It’s also possible to include topological relation information if extra efforts are taken to create a data structure representing topological relationships of building components. Textures can also be mapped to the building models using airborne and terrestrial imagery. The level of detail of the model depends on the feature detection algorithm from laser scanning points, the level of detail of the architectural plans and data integration method. Currently, the data pre-processing of architectural plans may need human interactions, whereas the reconstruction of roof, façade and indoor models and the integration of these models can be highly automated.

5 Preliminary Implementations 5.1

Registration of TLS Data and Floor Plans

In Wang and Sohn (2010), we propose a semantic and geometric information integrated point matching based method for co-registration of 3D TLS data and 2D floor plans, which demonstrates the feasibility to georeference floor plans by using TLS points.

268

L. Wang and G. Sohn

17

13 9

Corner Node Wall Window

44

1

Fig. 2 A sample floor plan and the corresponding semantically and geometrically parsed floor outline string

The main processes of the registration include: l l l

Extract building outlines from floor plans and TLS points respectively Find the corresponding points between the two outlines Calculate the coordinate transformation from floor plans to TLS data based on the corresponding points

First, as shown in Fig. 2, the outlines of the building are extracted from each floor plan. In our test, in order to make a simple and clear demonstration, we only use part of the floor plans of a big building. Normally, the floor plan outlines are polygons. Then, as shown in Fig. 3, a plane sweep technique based method is used to extract 2D building outlines from 3D TLS points. First, the original TLS points are segmented and façade planes are derived. The points belong to the walls can be extracted and projected onto the corresponding façade planes as shown in Fig. 3b. Then a level plane is used to scan and cut the facade planes bottom-up at a certain interval, for example 0.2 m; and at each vertical location, a TLS building outline can be extracted by intersecting the sweep plane with façade planes as shown in Fig. 3c. After finishing the bottom-up scan, a series of TLS building outlines are obtained. Because in practice all facades of a building can hardly be fully covered by TLS points, most generally a building is only partially covered by one or few scans. This means most of the time TLS outlines are lines. To make it a general algorithm, in our test we select TLS points from two single scans that cover part of the building facade. Therefore, the registration problem now is to find the corresponding points between the two kinds of outlines, which is a line-polygon matching problem. In order to reduce the difficulty to find the corresponding points and to transfer semantic information to TLS data, the outlines are semantically, geometrically and


269

Fig. 3 2D building outlines extraction from 3D TLS points (a) Original TLS points. (b) Segmented TLS points and sweep plane. (c) One of the TLS building outlines derived by intersecting sweep plane and the façade planes. (d) Segmented c

topologically parsed by encoding the invariant semantic and geometric context information to each inner point based on a defined shape context descriptor, which defines the invariant shape context information of an inner point in the outlines. As shown in Fig. 2, the floor plan outlines are encoded into floor outline strings by three steps: 1. Inner points are classified into node and corner. 2. Invariant semantic and geometric context information of an inner point such as point class, angle and length ratio between the two neighboring lines, and class of the two neighboring lines, are coded as the attributes of the node or corner. 3. Points are numbered and ordered in anticlockwise to represent their topological relationship. Points and line segments of TLS building outlines may have unknown value. But after registration, they will get the semantic information from their corresponding points and line segments and this information is useful for recognizing and reconstructing the façade features as presented in next section. Figure 3c shows one of the TLS building outlines. It needs to be further segmented into meaningful line segments. This can be done by projecting the laser scanning points within a strip of interval distance onto it. This way, line segments that have projected wall laser points are extracted (Fig. 3d). These line segments will have wall attribute.

270

L. Wang and G. Sohn × 104

1 4

4.7805

7 Corner

4.78

Node Wall Unknown

4.7795

4.779

4.7785

20 6.2005

6.2005

6.2006

6.2007

6.2007

6.2008

× 105

Fig. 4 Semantically and geometrically parsed TLS outline chain of Fig. 3d

Then using the similar coding method, TLS building outline are semantically, geometrically and topologically parsed into TLS outline chains as show in Fig. 4. Thus, the correspondence problem is converted to find the points from TLS outline chains (Fig. 4) and floor outline strings (Fig. 2) that have the similar attributes. A relaxation based coarse-to-fine matching process is used to find all the corresponding points. First, a similarity score formula is developed to find the points that have the same attributes. Then using these points as initial matching points, an iterative refinement is performed to find all the potential corresponding points. The vertical location where most correspondences are found for a floor plan outline is called the best matching position of this floor plan. The coordinate transformation parameters for this floor plan are calculated based on the corresponding points at the best matching position. Figure 5 shows a test result of the registration of TLS points and floor plans. The elevations of the floor plans are the heights of their best matching position. We assume that from the floor plan files we know which storey a floor plan corresponds to.

5.2 5.2.1

Integrated 3D Façade and Indoor Reconstruction Façade Feature Outline Generation

After registration, the horizontal extents and locations of façade features like windows and doors are known. Then a sweep line based method is used to find


271

Fig. 5 Registration result: floor building outlines are georeferenced and aligned with TLS points

Wall

Window

Vertical sweep line

Horizontal sweep line

Fig. 6 Sweep line based detection and generation of façade feature outline

the façade feature outlines. First, as shown in Fig. 6, along each façade feature line segment in the floor line string, a vertical sweep line is used to scan the projected laser scanning points horizontally to find the points on the upper and lower edges. At the same time, starting from the façade feature line segment a horizontal sweep line is used to scan upwards and downwards to find the points on the left and right edges. From the extracted edge points as shown in the right illustration of Fig. 6, edges can be derived by best line fitting and best curve fitting. Then these best fitting lines and curves are connected to generate the façade feature outlines. As can be seen, one advantage of this sweep line based method is that facade features with curved outlines can also be reconstructed.

272

L. Wang and G. Sohn

Fig. 7 Storey reconstruction (wall extrusion and windows integration)

5.2.2

Façade Feature Intrusion

Façade features, like windows and doors, normally are not on the same plane of the wall. The intrusion values can be obtained from 2D floor plans. However, they can also be automatically detected from the laser scanning points. For example, after generating the outline of a window, TLS points inside the outline can be extracted, among which some points may be on the frame of the window and some may be on the curtain if one exists. Then by analyzing the depths from the extracted points to the window plane, the intrusion value can be derived.

5.2.3

Storey Height Determination

In most cases, 2D floor plans do not provide storey height information. The storey heights can be obtained from other architectural drawings, for example elevation drawings. When the storey heights of a building are equal or can be assumed to be equal, especially when a building has regular or symmetric window structures between storeys, the storey heights can be automatically estimated by three steps: (1) Calculate the center points of the windows; (2) calculate the heights between the vertical neighboring two windows between two storey; (3) obtain the storey height by calculating the average height values between windows.

5.2.4

Storey Reconstruction

Using the obtained storey heights, floor plans can be extruded to form 3D presentation. Here, we only use the walls for demonstration. In our approach, the outer walls and the inner walls are extruded separately. We use the TLS outline chains for the reconstruction of outer walls to ensure that the outer walls of each storey are in the same façade plane. After the extrusion, the reconstructed façade features like windows are integrated with the external walls to form the storey model as shown in Fig. 7.


273

Fig. 8 Integrated 3D façade and indoor model

5.2.5

Final Test Building Model

Figure 8 shows the resulting integrated 3D building model, in which the façades are reconstructed mainly form TLS data with the assistance of the information from architectural plans and the indoor model is constructed using floor plans which are georeferenced by TLS data. Consequently, the façade model and the indoor model are integrated geometrically, semantically and seamlessly.

6 Conclusions In this paper, a framework for integrated full 3D building model reconstruction by fusing ALS data, TLS data and architectural plans is presented. The objective is to reconstruct 3D building models with complete roof, façade and indoor structure and semantic information in one workflow. As a proof-of-concept implementation of the proposed framework, an integrated 3D façade and indoor building model is reconstructed by fusing the information from TLS data and architectural plans. The test, although preliminary, has demonstrated the feasibility of the proposed framework. It also shows a promising way to bridge the gap in the data fusion between architectural engineering and construction community and GIS and geomatics engineering industry. In next step, further implementation will be performed by using more test datasets to improve the registration algorithm and the integrated façade and indoor model reconstruction. In order to improve the automation, new methods need to be developed to automatically estimate the storey height and distinguish floor plans of different levels. The future research will focus on the data integration and representation of 3D rooftop, facade and indoor models. The potential benefits to use other architectural drawings will also be investigated.

274

L. Wang and G. Sohn

References Becker S (2009) Generation and application of rules for quality dependent façade reconstruction. ISPRS Journal of Photogrammetry and Remote Sensing, vol. 64, no. 6, pp. 640–653 Brenner C (2000) Towards fully automatic generation of city models. International Archives of Photogrammetry, Remote Sensing and Spatial Information Science, vol. XXXIII, part B3, pp. 85–92 Horna S, Meneveauxa D, Damiandb G, Bertranda Y (2009) Consistency constraints and 3D building reconstruction. Computer-Aided Design, vol. 41, no. 1, pp. 13–27 Kada M, McKinley L (2009) 3D building reconstruction from lidar based on a cell decomposition approach. International Archives of Photogrammetry, Remote Sensing and Spatial Information Science, vol. XXXVIII, no. 3/W4, pp. 47–52 Lu T, Tai CL, Su F, Cai S (2005) A new recognition model for electronic architectural drawings. Computer-Aided Design, vol. 37, no. 10, pp. 1053–1069 Meijers M, Zlatanova S, Pfeifer N (2005) 3D Geo-information indoors: structuring for evacuation. In: Proceedings of Next Generation 3D City Models, Bonn, Germany, pp. 11–16 M€ uller P, Zeng G, Wonka P, Van Gool L (2007) Image-based procedural modeling of facades. In: Proceedings of ACM SIGGRAPH 2007/ACM Transactions on Graphics, vol. 26, no. 3, Article 85 Or SH, Wong KH, Yu YK, Chang MMY (2005) Highly automatic approach to architectural floor plan image understanding and model generation. In: Proceedings of Vision, Modeling and Visualization 2005, Erlangen, Germany, pp. 25–32 Oude Elberink S (2010) Acquisition of 3D topography. Ph.D. thesis, International Institute for Geo-information Science and Earth Observation, Twente University of Technology Pu S (2010) Knowledge based building facade reconstruction from laser point clouds and images. Ph.D. thesis, International Institute for Geo-information Science and Earth Observation, Twente University of Technology Ripperda N, Brenner C (2009) Application of a formal grammar to facade reconstruction in semiautomatic and automatic environments. In: Proceedings of 12th AGILE Conference on GIScience, Hannove, Germany Suveg I, Vosselman G (2004) Reconstruction of 3D building models from aerial images and maps. International Journal of Photogrammetry and Remote Sensing, vol. 58, no. (3–4), pp. 202–224 Wang L, Sohn G (2010) Automatic co-registration of terrestrial laser scanning data and 2D floor plan. In: Proceedings of Canadian Geomatics Conference 2010 and ISPRS COM I Symposium, Calgary, Canada, June 14–18, 2010 Yin X, Wonka P, Razdan A (2009) Generating 3D building models from architectural drawings: a survey. IEEE Computer Graphics and Applications, vol. 29, no. 1, pp. 20–30

Towards Semantic 3D City Modeling and Visual Explorations Qing Zhu, Junqiao Zhao, Zhiqiang Du, Yeting Zhang, Weiping Xu, Xiao Xie, Yulin Ding, Fei Wang, and Tingsong Wang

Abstract In recent years, the integration of semantics into 3D city models has become a consensus. The CityGML standard laid the foundation for the storage and application of semantics, which boosts the progress of semantic 3D city modeling. This paper reports an extended semantic model based on CityGML and its visual applications under the content of a three-dimensional GIS project of China. Firstly, concepts Room, Corridor and Stair are derived from concept Space which represents the similar concept of Room in CityGML. These concepts will benefit the application of indoor navigation. Geological model is also supported by this model, which enables the underground analysis. Secondly, a semi-automatic data integration tool is developed. The types of semantic concept are defined based on the Technical Specification for Three-Dimensional City Modeling of China which leads to an adaptive way to assign semantics into pure geometry. In order to better visualize the models enriched by semantics, two fundamental techniques, data reduction and selective representation are then introduced. It shows that semantics could not only help improve the performance of exploration tasks but also enhance the efficiency of spatial cognition. Finally, two exploration cases are presented, one is indoor navigation, the semantic model is used to extract the geometric path and a semantics enhanced navigation routine is used, which greatly enriches the connotation of ordinary navigation applications; the other is a unified profiler, in order to fill up the cross-section correctly, semantics are incorporated, which help ensure the topological and semantic consistency.

Q. Zhu, J. Zhao (*), Z. Du, Y. Zhang, W. Xu, X. Xie, Y. Ding, F. Wang, and T. Wang State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, People’s Republic of China e-mail: [email protected], [email protected], [email protected]. edu.cn, [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]


275

276

Q. Zhu et al.

1 Introduction 1.1

Semantics in 3D Modeling and Exploration

The accessibility of 3D city models (3DCMs) has grown unprecedentedly, success of Google Earth and Bing Map 3D brings us a new time with 3D Geo-information (Butler 2006). However, increasing professional applications give rise to needs of conceptual meanings beyond geometry since the pure appearance representation mainly focus on the photorealistic visualization while ignore a full comprehension of the data. Numbers of applications like urban planning and facility management, disaster management and personal navigation require additional information, i.e. classification and relationship of components, about the city objects given in a standardized representation (Kwan and Lee 2005). Therefore, the 3DCMs must incorporate the geometry and the semantics. CityGML is an international standard for the representation and exchange of semantic 3D city and landscape models, which not only represents the shape and graphical appearance of city models but specifically addresses the object semantics and the representation of the thematic properties, taxonomies and aggregations (Gr€oger and Pl€umer 2009). However, the current version of CityGML does not include underground features like geological model and underground infrastructures (Emga˚rd and Zlatanova 2008a). Moreover, implicit definition of Room in CityGML is not enough for the accurate geometrical path extraction in the indoor navigation. On the other hand, existing datasets are often produced lacking of semantics by using photogrammetric approach or CAD tools. An efficient way should be proposed to complement the thematic meanings of the geometry. This paper aims at the improvement of the CityGML model in the visual exploration practice of large urban scenes. The main contributions lie in following aspects: Firstly, geology model is supported in the thematic model; Secondly, Space is introduced in our building semantics and Room, Corridor, Stair that derived from Space are employed to facilitate indoor navigation; Thirdly, a semi-automatic semantic enrichment tool is developed to help enrich semantics to existing geometric models; Then, two basic visual exploration techniques in our platform are illustrated which utilize semantics to enhance the exploration performance; At last, two exploration cases are demonstrated to show the application of semantic model and visual exploration techniques. Our test bed is the digitalized 3D city models of the Wuhan city, China, which covers 8,600 km2, is consist of 119,051 buildings.

1.2

Related Work

Increasing 3D city modeling projects have been carried out in recent years. As one of the pioneers, the 3D Berlin project firstly integrates semantics into the 3DCMs produced by traditional photogrammetric procedures in order to facilitate the information

Towards Semantic 3D City Modeling and Visual Explorations

277

querying and spatial analysis (D€ ollner et al. 2006). At the same time, fundamental issues such as the consistency of geometry and topology as well as the coherent of geometry and semantics were extensively studied (Kolbe 2008; Kwan and Lee 2005), which laid the basis of CityGML and semantic modeling. Recently, several extensions of CityGML are also proposed, such as the integration of both above and underground features as well as temporal semantics of house properties (Emga˚rd and Zlatanova 2008a, b; Zhu and Hu 2010). On the other hand, visual exploration has been approved to be the most powerful tool to present and use the 3D spatial data. LandXplore in 3D Berlin project has become a standard package to visually interact with CityGML datasets (D€ollner et al. 2006), and many other researches on city visualization have proposed advanced features to better present 3DCMs (Fabritius et al. 2009). However, the full use of semantics is still lack in existing visual applications.

2 Semantic Modeling An “integrated 3D semantic model” is brought forward here, based on which an approach of “semantic enrichment” is discussed.

2.1

Integrated Model Based on CityGML

The integrated 3D semantic model adopts several concepts presented in CityGML but also comprise new developments which the current version of CityGML does not include. Figure 1 shows the structure of the thematic model, which is defined based on the Technical Specification for Three-Dimensional City Modeling of China (Nagel et al. 2009). An extended class Geology is added to support geological analysis applications, similar to (Emga˚rd and Zlatanova 2008a). The transportation here is specialized into express way, main road etc. Such subdivision makes the semantic enriching process more adaptive. In the following, the building model will be discussed in more detail to illustrate the general principle which is different from the definition in CityGML. A Building is described by optional attributes and constituents; optional attributes contain: function, usage, and class, measured height, number and so on. It is aggregated by two classes: IntBuilding and ExtBuilding, which facilitates the extracting of LOD3 model from the detailed model. Specifically, we borrow the abstract concept Space from IFC (Industry Foundation Classes) to represent the conceptual bounded space in a building (buildingSMART International 2010). Subclasses of Room, Stair, ElevatorRoom, and Corridor are derived from Space, which further define specified Spaces with typical composite pattern of Openings (Openings can be interpreted as doors, windows or entrances) This model is more appropriate to the application of indoor navigation because sometimes it is ambiguous to automatically extract

278

Q. Zhu et al. Branch_Road ParkingPlace MunicipalRoad

SemanticTree –

DEM TIN GRID

–

CountrysRoad

Building Residential_Building Commercial_Building Public_Building Industrial_Building Warehouse CityFurniture_Building Special_Building

–

TrunkRoad

Transportyard_Building Others CityFurniture ElectricPower_Pipeline Telecommunication_Pipeline Heat_Pipeline

–

–

Vegetation_Independence –

RoadsSquareSites ConstructionSites Specially_designatedLand

Rain_Pipeline Sewer_Pipeline Independence Transportation Express_way MainRoad MinorRoad

Vegetation_Coverage LandUse ResidentialLand Public_Facilities IndustrialLand WarehouseLand

Burningline Water-supplyline

–

HousesRoad Subway WaterBody WaterSurface Vegetation

–

GreenSpace Water_and_Miscellaneous Geology Stratum GeologicStructure GeologicBody GeologicalEngineering

Fig. 1 The thematic model

geometric path directly from the topological accessibility graph derived from Room and Opening of CityGML. For example, a stair with two entrances at the two ends clearly defines a path. However, a path extracted from the Stair in CityGML would be probably a vertical line segment as shown in Lee (2001), which is obviously not suitable for more precise navigation applications. In the following, definitions of the derived concepts are shown: l

Room: A Room is a specified Space which should have functions other than passing through, such as resting, working or entertaining etc. It should be bounded by WallSurfaces (Similar to the BoundarySurface in CityGML) like interior wall surfaces, ground surfaces, roof surfaces, Openings, and Furniture as well as other IntBuildingInstallations. A Room contains a label to record the WallSurfaces, by which the adjacent Space is connected through Openings. In the topological network, a Room is represented as a node.

Towards Semantic 3D City Modeling and Visual Explorations l

l

279

Stair: A Stair is a specified Space which contains a Stair and is bounded by WallSurfaces with at least two Openings, as shown in Fig. 2. A Stair contains a label to record the bounded WallSurfaces as well as Openings, by which the adjacent Space is connected. In the topological network, a Stair is represented by a node. Corridor: A Corridor is a specified Space which contains a passageway and is bounded by WallSurfaces with at least two Openings, as shown in Fig. 3. A Corridor contains a label to record the bounded WallSurfaces as well as Openings, by which the adjacent Space is connected. In the topological network, a Corridor is represented by a node.

The UML diagram of the building model is shown in Fig. 4. Via the definition, the model is more convenient and precise for constructing the geometrical route network automatically.

Fig. 2 A Stair

Fig. 3 A Corridor

Fig. 4 The UML diagram of building

280 Q. Zhu et al.


281

Fig. 5 The Geometry-Semantic coherent mapping

To record the Geometry-Semantic coherence, a flexible and extendable structure is needed, as illustrated in Fig. 5. Firstly, abstract thematic module is defined, from which hierarchy of semantics as well as topology could be derived. Then, semantic information is mapping to its corresponding geometrical objects like a solid or a surface. Under this framework, the geometry and appearance structure could be compactly designed for high performance visualization purpose and different domains of semantics can be mapped to the geometry model and be utilized in different applications in the meanwhile.

2.2

Semantic Enrichment

To seamlessly integrate heterogeneous data (particularly those without semantic information), it is necessary to incorporate or enrich the semantics for the geometries. While existing CityGML models could be imported and support well, for other models, however, it is hard to extract semantics from pure geometric models automatically. Therefore, a semi-automate tool is developed. As discussed in Sect. 2.1, the semantic concepts were already defined in thematic model. The assigning of semantic information and the building of semantic hierarchy turns out to be simply choosing the geometric components in top-down order (from exterior to interior) and indicating the corresponding semantic concepts. To improve the efficiency, an adaptive way is employed which could automatically refine the

282

Q. Zhu et al.

Fig. 6 (a) The imported geometric model; (b) mapping the entire building; (c) the component editing; (d) the component aggregation

selection range of semantic concepts according to the thematic model. After all the components are mapped, the semantic hierarchy is built at the same time. The integration procedure is illustrated in Fig. 6. Firstly, the entire model is mapped to its conceptual node of the thematic model, for example, Fig. 6b shows the building is mapped to the commercial building. Then, in the step termed component editing, the geometry-semantic coherence is built up by grouping geometry surfaces like wall and door etc. and mapping them to the semantic concepts derived from the abstract building, as illustrated in Fig. 6c. Finally, in the step of component aggregation, the mapped components are aggregated based on the thematic hierarchy, such as a closed Space bounded by Wallsurface would be aggregated into a Room or a Corridor, as illustrated in Fig. 6d. After the semantic enrichment, a semantic model is acquired, based on which various applications like conceptual querying and emergency path searching can be implemented.

3 Fundamental Approaches for Visual Exploration Three-dimensional visual exploration is one of the most important applications of 3D GIS, which not only help user investigate the highly detailed spatial data sets but also achieve an unexpected discovery of the hidden spatial knowledge


283

(MOHURD 2010). However, as the increasing of the accessibility of detailed 3D city models, real-time visualization of a complex city scene is still extremely challenging. On the other hand, users’ attention might be frequently diverted due to the unfocused representation, which would greatly decrease the efficiency of the spatial cognition. Given these reasons, we present two fundamental approaches for visual exploration: data reduction and selective representation.

3.1

Data Reduction Based on Semantics

A number of output sensitive technologies for visualization of massive scene had been proposed in recent years, which combine simplification, levels of detail, image-based rendering, visibility culling, as well as cache coherent layout methods. A good review is given in Gobbetti et al. (2008). However, researches on cognition and psychophysics suggest that human beings will always tend to neglect objects in the view fields which have little correlation with certain tasks, such as in navigation scenarios (Canosa 2009). Therefore, discarding or simplifying meaningless objects during exploration would further improve the performance of rendering but affect little of the cognition. In order to decide what to show and what to discard, criteria need to be set up to measure the importance of objects. For example, during indoor navigation, fine interior structures and surrounding components of the building are of great importance while terrain and outside buildings, which contribute little to the task, could be discarded; for fly over application, geology model and interior building parts could be discarded, and so forth. Figure 7 illustrates the contribution composition of a building. We provide several semantics related criteria to calculate the contribution value such as visual importance represented by the number of projection pixels, task correlation represented by the distance to the active camera path, etc. Vital semantics based tags are assigned to components in order to ensure that important components will not be discarded easily. These tags are integrated into the data dispatching and rendering pipeline, as shown in Fig. 8.

3.2

Selective Representation

Rendering of places of interest and analysis results must be given prominence in visual exploration of complex scenes. Selective representation we proposed emphasizes essential objects which users may pay more attention to while fading out others. The potential advantages would be that users will take considerable less time to complete a search and recognition task in comparison to normal representation.

284 Fig. 7 Illustration of the contribution composition

Fig. 8 The flow chart of data dispatching and rendering

Q. Zhu et al.


285

Since visual properties like color, light and transparency are some of the crucial factors influencing the visual perception of city scenes, a multi-mode visualization framework is employed, in which single object or a group of objects can be assigned with various draw model including hidden, wire-frame, highlight as well as different levels of transparence. The configuration of render modes are set up based on properties of certain tasks. Screenshots of underground utilities query application are shown in Fig. 9, for an example. In Fig. 9d, buildings above ground remain texture-mode, while the terrain surface is drawn in translucent-mode. High-lighted pipes perform significantly well in drawing users’ gaze. Therefore it could be easy to locate the corresponding position of pipes on the ground. In the implementation, we found that data reduction could successfully improve the performance of rendering and selective representation leads to easier interacting with complex city scenes. In some sense, our approach can be interpreted as a practical way to create an efficient representation of an urban environment, and, as a way of finding unique knowledge of a large cityscape. Possible areas of utilization of this technology are applications such as route finding, and spatial analysis applications.

Fig. 9 (a) The stratum and tunnel below ground; (b) the highlight of meaningful buildings; (c) alleviating occlusion by translucent; (d) revealing the relationship between pipes and above ground features

286

Q. Zhu et al.

4 Exploration Case 1: Semantics Enhanced Indoor Navigation 4.1

Overview

Almost all of the city emergency response scenarios need indoor navigation. However, it is always a labor intensive work to build path network inside a building manually. Fortunately, Lee proposed a method to extract the indoor road network based on connectivity graph of the building parts (MacEachren and Kraak 2001). We transmit his idea to our semantic models and improve the results by using pre-discussed Stair and Corridor concepts. A semantics enhanced navigation interface is also introduced.

4.2

Route Deducing: Automatic Extraction of Path Network

The automatic extraction of path network can be divided into two steps: network topology construction and geometric network construction. Network topology is a pure graph that represents the adjacency, connectivity and hierarchical relationships among the semantic nodes. Based on the definitions of the semantic model of the building, the network topology can be constructed automatically (Gr€oger and Pl€ umer 2009). A connection will be built between two Spaces if they can be connected by the Opening. In order to implement network-based analysis such as shortest path algorithms, the network topology needs to be complemented by geometric properties, which accurately represent the cost of connection. Lee presented a well-developed method, “Straight-MAT”, to identify line segments from a simple polygon based on combinatorial data model (CDM) which describes the topological relationships between 3D geographic entities (MacEachren and Kraak 2001), as shown in Fig. 10. However, Stairs in this method are represented as vertical lines, which might lead to unconvincing routing results. Moreover, the existing method will face difficulty in dealing with a square Corridor because the direction of the path line could not be uniquely defined. To partly solve this problem, we employ different routines to extract path lines from different derived Spaces. For example, “StraightMAT” method is used to extract path inside a Room while we directly extract path of a Stair or a Corridor by connecting Entrances, based on our semantic model. Results are shown in Fig. 11 by green lines.

4.3

Enhanced Route Navigation

Unlike the ordinary navigation applications, it’s important for indoor navigation to emphasis the key features for the user during navigation. With the help of semantic model, we could extract information such as the Room number and the name of the


287

q13 q12 q11

q14

n0

n2

q10 q15 q0 q4

q1

q9

q2 q3

q8 q7

n1

q5 q6

Fig. 10 Lee’s “Straight-MAT” method

Fig. 11 Constructed geometry path

building part passing by. Selective representation routines are also adopted to highlight the important features such as doors, Stairs and fade out the unrelated features such as other Building parts and Furniture. Some of the results are show as follows. Figure 12a and b shows the calculated shortest path, represented, from outside and inside the building. In Fig 12b, the important features such as Stairs, Corridors and Doors are highlighted while other Building parts are shown transparent. Figure 12c and d show the semantic tags for notifying features during navigation, which would enhance the impression of the route.

288

Q. Zhu et al.

Fig. 12 (a) The calculated path observed from outside; (b) the calculated path observed from inside; (c) the semantic tag indicates the Corridor on the second floor; (d) the semantic tag indicates the Room 204

5 Exploration Case 2: A Unified 3D Profiler 5.1

Overview

The unified 3D profiler, which could help users make good sense of urban architecture, underground infrastructure, internal structure and interrelationship of stratum, is an effective tool in providing visual hints and revealing the context and spatial relation of objects. However, since features in 3D city scene vary largely depending on geometry, topology and semantics. The need to ensure the topological and semantic consistency has a strong impact on the validity of profiling explorations and the analysis results.

5.2

Achieve Consistency for Topology and Semantics

Standard languages for 3DCMs such as CityGML provide “Solid” as the most important type to represent buildings, rooms, public infrastructure or other volumetric object in geometrical modeling. “Solid” here is described mathematically by rigid body, which is a bounded, regular and semi-analytical subset of R3 (Kolbe 2008). In boundary representation schemes, which are widely used in Geometrical Modeling,


289

Fig. 13 The extracted intersecting-line set of a geometrical model

CAD and GIS, solids are represented by their bounding surfaces, or topologically, a single, closed two-manifold. During profiling, each solid should be rebuilt in accordance with topology consistency. Otherwise, the ill-structured cross-section would prohibit not only the successive spatial analysis, but also the clear cognition of objects. On the other hand, for 3DCMs, marking, identification and affiliation descriptions of object type should be implemented with semantic consistency, which is the essential difference that distinguishes 3D city models from general geometrical models. In profiling, we cannot tell the meaning of cross sections in a reconstruction with inconsistent semantics. And it is also difficult to ensure query accuracy and descriptive validity of relationships among objects in this case. An example of profiling without considering topological and semantic consistency is given in Fig. 13. The intricate and complicated intersecting-lines are difficult to be handled in other applications. As a result, in order to ensure consistency of topology and semantics, implementation of profiling for buildings requires containing different semantic information derived from original semantic property as well as avoiding topological mistakes such as degeneration and punctuation. In addition, during profiling, whether to fill up cross-sections or just extracting cross-section lines should be judged according to the original semantic and topological information. Then, the generated surfaces should be mapped to semantic node in semantic hierarchy.

5.3

Profiling with Semantics

In the profiling analysis, objects in a city scene can be divided into three categories according to their topological descriptions, namely, models represented by open surfaces, models represented by strictly closed surfaces and models represented by a mixture of both open surfaces and closed surfaces. The first category contains terrain surface, which is the LOD0 model of CityGML. The second category consists of stratum in geology and models from LOD1 to LOD3 of CityGML as well as architecture components in BIM models,

290

Q. Zhu et al.

which are described by rigid bodies. The third category is mainly the LOD4 models of CityGML, in which building components such as walls, floors are expressed by open surfaces and rooms, IntBuildingInstallations and Furniture are expressed by solids, a good illustration of the differences between models of LOD4 and BIM models can be found in Gr€oger et al. (2008). During profiling, cross-sections should be treated separately, models belong to the first category only produce cross-section lines, and therefore, post processing is needless. For models belongs to the second category, cross-section must be filled by surfaces which can be determined based on the topology of the original model directly. However, to cope with Spaces in BIM models, inner features behind the cross-section such as bounding walls should be revealed so the generated surface should therefore be shown in translucent mode. For the models belong to the third category, which is the most complicated, different classes of surfaces should be

Fig. 14 The configuration of a unified profiler


291

distinguished based on the semantics so that we could fill the cross-section of a Space with translucent surfaces which contain the semantic information associated to the Space, while treat other cross sections as WallSurfaces which are filled by solid surfaces in order to close the gap between Spaces. The IntBuildingInstallations as well as Furniture are dealt with employing the same method used for the second category. The main steps of the profiling are as following: l

l

l

l

Firstly, objects in the scene are classified into different categories according to the criteria discussed above. Then, intersection lines are extracted between the cross-section surface and intersected surfaces. Delaunay triangulation routine is adopted to re-mesh the cross-section. Thirdly, a BSP tree is built and semantic information, especially the Space, is employed to classify the new faces. At last, the cross-section is filled up according to semantics and is mapped to the semantic hierarchy which ensures that new elements inheriting semantic information correctly from the original model.

5.4

Results

Figure 14 shows the configuration of a profiler and Fig. 15 shows a built crosssection surface. Different types of cross-section surface including planes and curved surfaces are supported. And what is the most important, the participants of

Fig. 15 A threefold cross-section surface generated by the profiler

292

Q. Zhu et al.

the profiling application, such as which kind of feature should be profiled, could also be defined. Examples of profiling with typical types of models in city scene are given in Fig. 16. The top shows a LOD4 model, with geometry and texture profiled correspondingly. The restored cross-section is displayed transparently so as to show inner features. The middle shows a detailed architecture model without Space. All the components are filled up by solid surface with the same material after profiling. For example, the gray part on the first floor shows a cross-section of a Buddha statue, which was filled up by the right texture. The bottom shows a geology model

Fig. 16 Profiling results. Top: a LOD3 building; middle: an IFC building; bottom: geology model


293

Fig. 17 An example of user-defined profiler based on semantics

represented as a layer-cake. The profiling is executed in each stratum, which is properly filled up. An example of user-defined profiler is depicted in Fig. 17, in which Furniture “sofa” is set to be included in the analysis, while floor and Furniture “table” are excluded. In this exploration case, it is demonstrated that the unified profiler based on the topological and semantic consistency could provide a dimensional reduction technique, which would probably be one of the basic analyzing technologies in 3D GIS.

6 Conclusion This paper presents a semantic model based on CityGML, which is extended to support geological model and specified Spaces such as Stair and Corridor for indoor navigation. An integration tool is also proposed for enriching existing geometric models with semantics. Besides, two visualization techniques based on semantics are introduced which aim at improving the performance of visual exploration. All the above laid the foundation of our 3DGIS platform, in which the geometry and material data structure is designed mainly for high performance visualization but can be flexibly mapped to the semantic hierarchy. Based on the platform, we achieved the semantics enhanced indoor routing and unified profiling of the complex city scenes. Currently the ongoing project is developed to support most of the commonly used data types and extend the storage of the massive city dataset to the commercial database. Moreover, dynamic features like fluid and clouds are expected to integrate in the future. Acknowledgements The research for this paper was funded by the National Basic Research Program of China (No. 2010CB731800), National Natural Science Foundation of China (40871212 and 40701144) as well as the National High Technology Research and Development Program of China (No. 2008AA121600).

294

Q. Zhu et al.

References buildingSMART International (2010) Ifc2x4 release candidate 1, http://www.iai-tech.org/groups/ msg-members/news/ifc2x4-release-candidate-1-available, 2 September 2010 Butler D (2006) Virtual globes: the web-wide world. Nature 439 (7078):776–778 Canosa RL (2009) Real-world vision: selective perception and task. ACM Transactions on Applied Perception (TAP) 6(2):1–34 D€ ollner J, Kolbe TH, Liecke F, Sgouros T, Teichmann K (2006) The virtual 3D city model of Berlin-managing, integrating, and communicating complex urban information. In: Proceedings of the 25th International Symposium on Urban Data Management. Aalborg, Denmark, pp. 15–17 Emga˚rd L, Zlatanova S (2008) Design of an integrated 3D information model. In: Coors, P et al (eds) Urban and Regional Data Management. Taylor & Francis Group, London, pp. 143–156 Emga˚rd L, Zlatanova S (2008) Implementation alternatives for an integrated 3D information model. In: van Oosterom, P et al. (eds) Advances in 3D Geoinformation Systems, Lecture Notes in Geoinformation and Cartography. Springer, Berlin Heidelberg, pp. 313–329 Fabritius G, Kraßnigg J, Krecklau L, Manthei C, Hornung A, Habbecke M, Kobbelt L (2009) City virtualization. In: The 5th Workshop Virtuelle und Erweiterte Realit€at der GI-Fachgruppe VR/ AR. Magdeburg Gobbetti E, Kasik D, Yoon S (2008) Technical strategies for massive model visualization. In: Proceedings of the ACM symposium on Solid and physical modeling. Stony Brook, New York, pp. 405–415 Gr€oger G, Pl€umer L (2009) How to achieve consistency for 3D city models. GeoInformatica. doi:10.1007/s10707-009-0091-6 Gr€ oger G, Kolbe TH, Czerwinski A, Nagel C (2008) OpenGIS city geography markup language (CityGML), Implementation Specification, Version 1.0.0, Implementation Specification, OGC Doc. No. 08-007r1 Kolbe TH (2008) Representing and exchanging 3D city models with CityGML. In: Zlatanova S, Lee J (eds) 3D Geo-Information Sciences. Springer, Berlin Heidelberg, pp. 15–31 Kwan MP, Lee J (2005) Emergency response after 9/11: the potential of real-time 3D GIS for quick emergency response in micro-spatial environments. Computers, Environment and Urban Systems 29(2):93–113 Lee J (2001) A 3D data model for representing topological relationships between spatial entities in built-environments, Ph.D. thesis, The Ohio State University, Columbus, Ohio MacEachren AM, Kraak M (2001) Research challenges in geovisualization. Cartography and Geographic Information Science 28:3–12 Ministry of Housing and Urban-Rural Development of the People’s Republic of China (MOHURD) (2010) Technical specification for three dimensional city modeling of China, http://vrlab.whu.edu.cn/863/News/20100513.htm (in Chinese) Nagel C, Stadler A, Kolbe TH (2009) Conceptual requirements for the automatic reconstruction of building information models from uninterpreted 3D models. In: Kolbe TH, Zhang H, Zlatanova S (eds) GeoWeb 2009 Academic Track – Cityscapes. Vancouver, BC, Canada Zhu Q, Hu MY (2010) Semantics-based 3D dynamic hierarchical house property model. International Journal of Geographical Information Science 24(2):165–188

Advances in 3D Geo-Information Sciences

Advances in 3D Geoinformation Systems (Lecture Notes in Geoinformation and Cartography)

Recent Advances in the 3D Physiological Human

Advances in Digital Terrain Analysis (Lecture Notes in Geoinformation and Cartography) (Lecture Notes in Geoinformation and Cartography)

Innovations in 3D Geo Information Systems (Lecture Notes in Geoinformation and Cartography)

Advances in Pharmaceutical Sciences, Volume 7

Advances in Pharmaceutical Sciences, Volume 7

Advanced Geoinformation Science

Vision in 3D Environments

Albtraum in 3D

3D Spectroscopy in Astronomy

Calculus in 3D (Draft)

Albtraum in 3D

Advances in Configural Frequency Analysis (Methodology In The Social Sciences)

Advances in Configural Frequency Analysis (Methodology In The Social Sciences)

Advances in Stromatolite Geobiology (Lecture Notes in Earth Sciences 131)

Learning in 3D

Ideas in Chemistry and Molecular Sciences: Advances in Synthetic Chemistry

Advances in Research in Karst Media (Environmental Earth Sciences)

3D Rendering in Computer Graphics

Advances in Interdisciplinary Applied Discrete Mathematics (Interdisciplinary Mathematical Sciences)

Chemically Modified Electrodes (Advances in Electrochemical Sciences and Engineering)

Innovations and Advances in Computer Sciences and Engineering

Advances in Computer and Information Sciences and Engineering

Achievement Testing: Recent Advances (Quantitative Applications in the Social Sciences)

Advances in Fluid Mechanics VIII (Wit Transactions on Engineering Sciences)

3d-ангел

Geographic Hypermedia (Lecture Notes in Geoinformation and Cartography)

3D Echocardiography

Advances in 3D Geo-Information Sciences

Advances in 3D Geoinformation Systems (Lecture Notes in Geoinformation and Cartography)

Recent Advances in the 3D Physiological Human

Advances in Digital Terrain Analysis (Lecture Notes in Geoinformation and Cartography) (Lecture Notes in Geoinformation and Cartography)

Innovations in 3D Geo Information Systems (Lecture Notes in Geoinformation and Cartography)

Advances in Pharmaceutical Sciences, Volume 7

Advances in Pharmaceutical Sciences, Volume 7

Advanced Geoinformation Science

Vision in 3D Environments

Albtraum in 3D

3D Spectroscopy in Astronomy

Calculus in 3D (Draft)

Albtraum in 3D

Advances in Configural Frequency Analysis (Methodology In The Social Sciences)

Advances in Configural Frequency Analysis (Methodology In The Social Sciences)

Advances in Stromatolite Geobiology (Lecture Notes in Earth Sciences 131)

Learning in 3D

Ideas in Chemistry and Molecular Sciences: Advances in Synthetic Chemistry

Advances in Research in Karst Media (Environmental Earth Sciences)

3D Rendering in Computer Graphics

Advances in Interdisciplinary Applied Discrete Mathematics (Interdisciplinary Mathematical Sciences)

Chemically Modified Electrodes (Advances in Electrochemical Sciences and Engineering)

Innovations and Advances in Computer Sciences and Engineering

Advances in Computer and Information Sciences and Engineering

Achievement Testing: Recent Advances (Quantitative Applications in the Social Sciences)

Advances in Fluid Mechanics VIII (Wit Transactions on Engineering Sciences)

3d-ангел

Geographic Hypermedia (Lecture Notes in Geoinformation and Cartography)

3D Echocardiography

Recommend Documents