EXTRACTION OF PLANES CONSISTING OF BUILDINGS FOR MODELING USING CLOSE-RANGE PHOTOGRAMMETRY

摘要

In recent years, the demand of three-dimensional model of buildings in cultural heritage and urban areas is growing rapidly. In dense urban areas, close-range digital photogrammetry is appropriate in terms of the cost and portability. However, in dense urban areas, the orientation may fail especially in case with narrow streets and many obstacles. In addition, it is quite time-consuming to manually select passpoints necessary for building modeling. Therefore, the objective of this study is to examine the preprocessing for the efficient modeling of the buildings in dense urban areas using digital photogrammetry. Automatic extraction of planes consisting of a building can help users in modeling. In the proposed approach, the planes are extracted using the three-dimensional coordinate and color data of points on buildings. First, relative orientation is conducted and a large number of passpoints are generated using SIFT (Scale-Invariant Feature Transform). Then, the passpoints are automatically classified into several independent planes. Finally, the boundary of the plane is determined through region-growing algorithm by referring to the brightness of images. The plane classification prior to the region-growing can merge the shadowed regions with non-shadowed regions, and reduce the shadow effect. In the experiment, images were obtained in Hanoi Old Quarter. As a result, it was found that three or four planes, walls and commercial signs, were successfully extracted from these images by using the proposed method. On the other hand, the other planes were not extracted because the planes had been divided by the obstacles or the patterns on the planes were intricate. In future, it is necessary to consider how to improve the accuracy of the extraction using the three-dimensional coordinate data instead of the color data.