Automatic generalization of digital building models

For interactive 3D visualization of a cyber city, the efficiency of browsing is highly depended on the number of triangles and textures to be rendered. Many algorithms regarding to terrain simplification has been discussed in the field of computer vision. In this study, we will focus on the generalization of 3D building models. In the paper, an automatic approach to generate three levels-of-detail building models is proposed. The most detailed one is the polyhedral building model that is generated using the SPLIT-MERGE-SHAPE algorithm. Buildings with gable roof, flat roof, with regular or irregular ground plan could be described. The second level is called the prismatic building model that has a flat roof only but remaining its elaborate ground plan. In which, all connected building primitives are merged into one building with the most appropriate height. The third level of building model is called the quadrilateral building model that is generalized from the prismatic building model using the dynamic-piping technique. All levels of generalization could be performed in an automatic way. Experimental results demonstrate that the proposed method is capable of generating three discrete LODs of 3D building models automatically and reduce the total of triangles effectively.