| dc.description.abstract | Three dimensional (3D) building model is one of the most important components in the construction and representation of urban scenes. In application systems, the vast amount of building data may lead to lower efficiency. For interactive visualization and real-time rendering in cyber city applications, Level of Detail (LOD) technique is a common approach to reduce unnecessary computations. LOD technique can generate different details of data and set appropriate levels according to viewing parameters to enhance visualization system performance.
This research proposes a semi-automatic approach for the generalization of 3D polyhedral building models to formulate LOD representations of complicated buildings for high performance visualization of cyber city. The idea is to apply generalization in 2D orthographic views of 3D building models. First, (geometric) structure analysis is conducted to compute shape complexity for determining the number of levels to generalize and to detect non-planar roof structures. For building models to generalize, orthographic views along different directions of the 3D building geometries are generated. To construct the topology of orthographic views, rasterization techniques and modified quasi-Target Defined Ground Operator (quasi-TDGO) are used in pre-processing of orthographic views. By calculating the lengths and angles with the topology, the edge regularization and U-, Z-, L-structures elimination are employed to generalize 2D orthographic views. Simplified 3D models are reconstructed from generalized 2D orthographic views with Boolean-based algorithms. For special structures of building models, such as buildings with non-planar roof structures, courtyards, and curved elements, additional processes are developed to simplify them but maintain their characteristics. Finally, these special structures are combined with reconstructed façades to complete the generalization of 3D building models.
Two groups of complex building models in National Central University (NCU) campus and the Xinyi District of Taipei city are used to test the developed building generalization algorithms. Experimental results of single building models demonstrate that the proposed method can effectively simplify 3D building geometries with generalized 2D orthographic views of building models. For building groups in large areas, the proposed method is effective in decreasing the number of points and polygons of building groups in different levels. In addition, the experimental results of building models with special structures are also validated, proving that the proposed approach can preserve important visual characteristics of complicated building models.
| en_US |