Texture-Cognition-Based 3D Building Model Generalization

Abstract

Three-dimensional (3D) building models have been widely used in the fields of urban planning, navigation and virtual geographic environments. These models incorporate many details to address the complexities of urban environments. Level-of-detail (LOD) technology is commonly used to model progressive transmission and visualization. These detailed groups of models can be replaced by a single model using generalization. In this paper, the texture features are first introduced into the generalization process, and a self-organizing mapping (SOM)-based algorithm is used for texture classification. In addition, a new cognition-based hierarchical algorithm is proposed for model-group clustering. First, a constrained Delaunay triangulation (CDT) is constructed using the footprints of building models that are segmented by a road network, and a preliminary proximity graph is extracted from the CDT by visibility analysis. Second, the graph is further segmented by the texture–feature and landmark models. Third, a minimum support tree (MST) is created from the segmented graph, and the final groups are obtained by linear detection and discrete-model conflation. Finally, these groups are conflated using small-triangle removal while preserving the original textures. The experimental results demonstrate the effectiveness of this algorithm.

Keywords

• 3D building generalization
• texture
• SOM
• cognition
• constrained Delaunay triangulation