|dc.description.abstract||As the virtual reality(VR) technology advances, geographic information systems(3D GIS) have been evolved from 2D to 3D. Building model has always been a major part in the applications of 3D GIS. Numerous studies have been working toward the generation of photorealistic building models in automated or semi-automated ways. Façade texture mapping that combines terrestrial laser scanning(TLS) data, close range images is proposed in this research. To strike a balance between reality and visualization, image occlusion detection and texture analysis is used in the process.
There are four main steps in the proposed method: (1) data registration, (2) forescene elements determination, (3) image occlusion detection and compensation, and (4) façade patch selection and texture mapping. First, a building model, close range images, and TLS data are registered in a unified coordinate system. Then, the forescene occlusion areas are detected by spatial analysis of the TLS point clouds. Forescene elements are backprojected onto images, and we compensate occluded parts from other angle images to create a large number of candidate patches. To recognize similar façade texture, we start from two different aspects, namely, façade patch geometric shape and façade patch texture. Compactness calculation and GLCM analysis are applied on these two characteristics individually. Texture characteristic and intensity value are considered in the analysis of repetitive texture patches to find the optimal façade texture. After the analysis, the optimal façade texture is selected by statistical clustering, accordingly. The experimental results show that the proposed mapping method in this research consider both aesthetic and authentic, and the process is applicable.||en_US|