טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
Ph.D Thesis
Ph.D StudentAlexey Noskov
Subject3D Generalization of Urban Environment
DepartmentDepartment of Civil and Environmental Engineering
Supervisor Professor Emeritus Doytsher Yerach
Full Thesis textFull thesis text - English Version


Abstract

Traditionally, generalization in general - and automatic generalization in particular - is a complex topic.  The 3D display of urban scenes is based on 3D urban models, which contain objects such as buildings, trees, surfaces, etc., all with specific geometric and semantic properties.  A 3D urban model is a computer reflection of the real urban environment.  The growing demand for developing and using 3D urban models has been expounded in many papers.  Many applications (such as tourism, geo-simulations, navigation, architecture etc.) could be improved by using 3D urban environment visualization.  However, the use of 3D models is very limited due to the inadequate level of 3D cartography in general, and 3D cartographic generalization in particular.

            Our approach to 3D generalization has been developed and tested on two city areas: Trento (Italy) and Haifa (Israel).  The approach could be applied for block building models.  The central idea is based on rectilinear grid simplification. 

            Source building objects were clustered into quarters.  In order to prepare different degrees of generalization, the quarters were grouped.  The groups of quarters were calculated for multiple zoom levels by applying a routing-based approach.

            In order to calculate quarter groups for various zoom level of the Haifa dataset, an OpenStreetMap road layer was used.  The layer was integrated with a building dataset using a developed linear approach.  Two approaches to road data integration have been developed.  First approach is based on segmentation principle.  It allows the user to define correspondent segments of polylines.  Disadvantages of the segmentation based approach are slow processing and many false positive cases which are difficult to detect automatically.  The second, linear based approach uses the triangulation techniques and set of topological tests.  This approach solves the raised problems.  It has been tested on various datasets.  The results are satisfactory; it was statistically proved.

            Initial approach to generalization of building is based on rasterization of 2D building footprints.  The generalized buildings’ footprints were extruded.  A set of information measurements was developed to evaluate the quality of the generalized layers: coordinate digit density, entropy of Voronoi polygons, and entropy of Voronoi areas.  A special parameter which allows the user to automatically define an optimal degree of generalization of buildings was developed.  The information measures are actively used in the final approach to 3D generalization of buildings for defining the optimal level of generalization.

            A rectilinear grid generalization of 3D building models was implemented.  Different degrees of generalization were calculated for different groups of building at various zoom levels.  In order to define an optimal generalization degree of quarter groups, voxel coordinate density entropy was calculated.  This parameter is based on coordinate digit density function used for evaluating a quality of 2D building footprint generalization.

            Resulting 3D perspectives of Haifa and Trento were calculated.  A set of trigonometric equations allows the user to define the visibility of a quarter and an optimal degree of generalization.  From the 3D perspectives review it could be concluded that the results are satisfactory.