Automated Building Extraction from Digital Stereo Imagery

Abstract

One of the most critical steps in 3D GIS database establishment and revision as well as automated digital photogrammetry is to capture the 3D man-made structures from stereo images automatically and accurately. The key problem for fulfilling such a task is to resolve the correspondence problem between image pairs. The most challenging problems exist in the large scale imagery of dense urban or suburban areas where the structures are very complex, irregular in shape and quite different in size. In this paper, we address an reliable and effective system aiming at reconstructing the ground surface and extracting buildings with relative disparity from the complex aerial imagery in urban areas. Four building hypothesis extraction models are addressed. Experimental results are demonstrated at the end of the paper.