From Cadastres to Urban Environments for 3D Geomarketing Martin Hachet and Pascal Guitton Abstract-- This paper presents tools performing automatic generation of urban environments for 3D geomarketing. Geomarketing aims at the visualization of strategic information (financial, sociological ) on geographical supports for decision making processes. The use of cadastral files allows to quickly generate the urban models while keeping direct links between the firms and their locations in the virtual environment. The 3D models are generated in an optimized way, enabling real time interaction with the environment using standard PCs. Index Terms-- 3D Models, Cadastres, Geomarketing, Urban Environments, Virtual Reality. I. INTRODUCTION IRTUAL Reality (VR) methodology has been primarily Vused for "real" data manipulation: planes, buildings, towns, human bodies... Then, some new applications have been developed dealing with abstract information: scientific, medical, financial, commercial... ; mainly to obtain a better understanding of complex data for decision making [7]. Geomarketing aims at the visualization of strategic information (financial, sociological ) on geographical supports for decision making processes. The size and complexity of the geomarketing data to be analyze led us to use the power of VR, for the visualization of 3D models as well as the interaction with them [1] [4] [10]. The first step of our work consists on generating 3D models of urban environments the user will be able to interact with in order to accomplish several geomarketing tasks for decision making processes. For example, he/she has to reply to questions like : "Where is the best place in town to open a new mexican restaurant?" The decision depends on constraints for the position such as : - being visible from both individual and collective transports, - being easily accessible (parkings, bus stops ), - being in an area with few exotic restaurants So, we need strong links between geographical supports and strategic data. For example, we must be able to easily position the representation of a type of restaurant at the place where the restaurant is located in the urban environment. Furthermore, we have to keep in mind that numerous updates have to be taken into account in our 3D models because of the constant evolution of the cities 1. For these reasons, our models are built from cadastral information regularly updated by the Town Halls. We use a standard format (MID-MIF from MapInfo [9]). This format is currently used by Geographical Information System (GIS), but we will see that existing GIS systems do not provide interaction with 3D models. The MIF files contain the shape of each building and the MID files contain some "high-level" information like the names of the companies or its identifier. These files are adapted to geomarketing, but are not adapted to 3D. To overcome this problem, we established rules to generate 3D models from these files. As we develop this methodology in order to create a real application software, we have to deal with some strong constraints as for example : automation of the urban environment construction, real time rendering, use of standard PC's to enable large diffusion and, finally, a minimum construction quality in order to allow a town recognition by the users. In Section 2, we study the existing works and we present the constraints of our project. In section 3, we describe the building methods of urban environments, and finally we give some examples and results in section 4. II. PREVIOUS WORKS AND CONSTRAINTS A. GIS and Urban Environment Our work is included in the GIS area, systems dealing with storage and manipulation of geographically organized information [3]. 3D GIS generally deal with Digital Elevation Models, but they don t deal with urban environments which are the most useful territories for marketing studies. Some of them propose multi-view displays, for example cartographic data, photography and user information. But, to our knowledge, none of the existing systems performs interactive image synthesis while mixing geographical and strategic data. The most classical way to generate urban environments is the use of aerial photographies [5][6] providing very precise models, but the required data is very heavy to obtain. These acquisition method can t be used for geomarketing because updates are hard to support. Starting from the fact that our application context does not need photographic precision, the generation of 3D models from aerial photographies is not a Martin Hachet and Pascal Guitton are with the LaBRI-Université Bordeaux I, 351 cours de la Libération, 33405 Talence Cedex, France (telephone: 33 556 846 900, e-mail: [hachet,guitton]@ labri.fr). 1 For example in Paris, 27000 modifcations are performed every year in the company list.
good idea. B. Constraints Besides the need to have a strong link between the strategic data and the 3D models, our geomarketing context induces some strong constraints : 1) Automatic and Generic Construction The application fits into an industrial process allowing the creation of urban models from any town with a minimum number of human operations. Moreover, updates in the database due to modifications in the real urban environment need to be performed easily and quickly. 2) Real Time Rendering To obtain interactive frame rates for virtual reality applications (about 20 frames/s), we have to create very "light" 3D models (in terms of polygons). 3) Standard PC The final application must run on "classical" workstations as PC's used by a large number of users. So, we have to optimize many parts of the execution process. 4) Town recognition The user must be able to recognize the real town when investigating the associated virtual environment. Our work is based on the set of components defined by Lynch [8]. generated as follows. On a first step, all the buildings are generated. On a second step, some of them are modified (churches, large buildings). Finally, elements of decor are added (trees, river ). Cadastral files give a very precise definition of the ground surface of the buildings (with an error inferior to one centimeter). The use of a naive algorithm for the elevation of the buildings would generate a large number of polygons per building, making it difficult to exploit the 3D model in real time. For example, the average building in Bordeaux consists of ten sides. A solution can be to simplify the shape of the ground surface of the buildings before performing an extrusion. Assuming that the set of vertices of the ground surface is { P i, 0 i n }, n being the number of vertices, the point P i is deleted if P i 1Pi + 1 ( Pi 1P i + Pi Pi + 1) < ε In this way, the number of sides to be generated can be decreased while keeping the general shape of the building (Fig. 2). III. GENERATION OF URBAN ENVIRONMENTS The given data to build our models are the delimitation of the buildings, the parcels, the blocks, the sections and the cities (Fig. 1). Each building belongs to a parcel. A block is a group of parcels, which is not cut by a street. A city is split into sections, which are split into blocks. Moreover, for some buildings the height in number of floors as well as its year of construction are known. block limit parcel building Fig. 2: Simplification of a building s ground surface using a distance oriented algorithm. However, this technique can't be used for buildings situated in a dense area because they have to be considered in their context. Individual simplifications can induce visual discontinuity as shown in Fig. 3. buildings street Fig. 3: Simplifying buildings individually induces visual discontinuity. Fig. 1: input data We deal with old European cities and therefore we can't use algorithms based on a regular topographical structure, as we were able to do with North-American cities. We are developing tools permitting the generation of 3D models of any city, as complex as it can be. Urban environments are Moreover, the position of a building relative to the street offers information that can be used to add a meaning to our virtual urban environments. By knowing the side(s) of the buildings facing the street, roofs can be designed with a logical orientation, and different specific textures for the different sides can be associated. Building cities from the streets permits to save a huge number of polygons. Effectively, we will see that the buildings can be generated from one side, the one facing the street, with a few numbers of polygons. The simplification only induces few losses concerning the visualization as the user generally knows a street only by the
facades it is composed of. He does not know the inside of the block of houses which can be complex (see Fig. 1). Cadastral information describes the buildings individually. The building sides facing the street have to be found. The given definition of the blocks we have is administrative and not geographical (i.e. a block is not the contour of a set of buildings). The contour of each block of houses has first to be found, this is done by merging the parcels. The second step consists on testing if the sides of the buildings belong to the contour. The only points belonging to the contour are kept. The next step consists on having a simplified ground surface by finding the back points. The first method we implemented was based on median axis decomposition as shown in figure 4a. It consists on allocating a space to each building on which a back projection is possible. This technique sometimes does not produce satisfying results (Fig. 4b, 4c). forward point, depending on the distance separating it from the others. - If the building has four points on the contour of the block with two parallel lines, we suppose that it corresponds to a building situated over both sides of the block. We generate the 3D model from these four points. - In all other cases, the real ground surface is used to generate the building. In a huge majority of cases, buildings correspond to one of the particular cases (about 90% in a dense area). A simplified model can be generated while keeping the general aspect of the real city (Fig 6). a b c Fig. 4: Use of the median axis to find the back points (a). The real shapes (b) can be bent (c) Our solution consists on applying heuristics to obtain a new ground surface composed of four points for the building facing the street as follows: - If the building has two points A and B on the contour of the block, we choose the third one between the point coming before A and the point coming after B in order to have the longer segment. Let C be this point. The fourth point is given by the intersection between the line parallel to (AB) passing through C and the line perpendicular to (AB) passing through the point opposite to C (Fig. 5). Figure 6: 2D and 3D view of a block of houses From the simplified ground surface of the buildings and from the number of floors they have, 3D description files can be generated. These 3D description files can be converted into VRML files or directly used by an application by means of the primitives of the chosen toolkit (OpenInventor, WorldToolKit ). The structure of the description files corresponds to a geographically organized scene graph (Fig. 7). C city section block buildings A Fig. 5: the exact ground surface of a building and its simplified one - If the building has three points on the contour of the block (following themselves), we suppose that it is a corner house. We choose the fourth point as the previous or the B Figure 7: Scene graph organization of the city Textures are chosen from a database, depending on the side and the size of the building. For the Bordeaux model, about twenty different textures are used. They correspond to pictures of real Bordeaux facades. Therefore the texture of a given building is not exact, but in a geomarketing context, we don't need an exact representation of each building. Once all buildings have been automatically generated, a human operation is needed to modify the buildings that are not visually satisfying. This can be performed by simply changing a texture, or by changing the whole building. We developed an
interface that permits to easily modifying the model online. The buildings that have to be changed are the ones corresponding to the landmarks of Lynch's cognitive maps (churches, huge buildings ). They are essential on the model because it is in a large part thanks to them that a user will be able to recognize a city. We developed tools that automatically generate a church from four points. Models coming from image based modeling [2] can also be integrated. In a last step, according to the data we have, different elements can be generated, such as trees, streets, and rivers. IV. RESULTS AND USE OF MODELS We obtain visually satisfying models that allow an interactive navigation on standard PCs. Fig. 8 and 9 show the result of a Bordeaux 4 km² surface generation. This surface contains 8712 generic buildings, 10 individual buildings, some hundreds of trees and streets, one river as well as a sky representation. We also deal with a dynamic model of the future tramway of Bordeaux. We obtain a model composed of 108040 polygons, in which it is possible to navigate fluidly by using WorldToolKit with an 800 MHz PC using a GeForce 2Mx graphics board. Fig. 8: General view of 3D model of Bordeaux Fig. 9: View of 3D model of Bordeaux
Each graphical element is referenced and the strategic data frame is the same as the one of the urban model. The abstract data is then integrated immediately. For example, if we have a list of existing drugstores (i.e. a list of building identifiers corresponding to the drugstores) it is very easy to localize them in the 3D model by displaying them with a green color or by pointing to them with specific pointers. Fig.10 shows an example of a user request enabling the visualization of the influence zone of two drugstores. We associated the shop information (identifier, location, and size ) to graphical representations: spot lights illuminating areas. Fig. 10 : Visualization of the influence zones of two drugstores Our work shows that it is possible to use 3D and real time interactions for geomarketing using standard PCs. From cadastres we are able to generate urban environments which are not true representations, but enough likeness to be used for geomarketing applications. Finding new 3D metaphors for the representation of the abstract information constitutes a new axis for research. ACKNOWLEDGMENT This work has been done in collaboration with Cartegie, a consulting service specialized in geomarketing; they provided the input data. [4] P. Guitton and E. Zeghers, Géomarketing 3D, Proceedings of MICAD 2000, Paris, mars 2000. [5] ISTAR information available at http://www.istar.fr [6] IVT information available at http://www.ivt.fr [7] R. S. Kalawsky, The science of virtual reality and virtual environments, Addison-Wesley, 1993. [8] K.Lynch, The image of the city. Cambridge: M.I.T. Press 1960. [9] Mapinfo information available at http://www.mapinfo.com [10] B. Robertson, Biz Viz gets real, Computer Graphics World, pp. 29-34, avril 1999. REFERENCES [1] D. N. Chorafias and H. Steinmann, Virtual Reality : Practical Applications in Business and Industry, Prentice Hall, 1995. [2] P. E. Debevec, C. J. Taylor and J. MALIK, Modeling and rendering architecture from photographs: A hybrid and image-based approch. SIGGRAPH 1996. [3] GIS information available at http://apollo.ogis.state.me.us/links/gislinks.asp