| Under the powerful motivation of the human society, remote sensing technology has gained a great deal of concerns and rapidly progress. Optical remote sensing data can describe the ground objects more precisely, so there is a great deal of application values in the related application research field, especially in the target recognition of buildings, and its three-dimensional reconstruction. As the core issue of building three-dimensional reconstruction, building feature extraction and stereo matching is of significant importance. Focus on the solution of these intractable problems, the thesis makes further research on building extraction and stereo matching technology.Firstly, related foundational theories of building extraction and stereo matching are discussed in this thesis. A brief introduction of building features in remote sensing images and styles of building modeling is given. Then classification of stereo matching is particularly introduced, and furthermore, restrain problems as the most important issue in stereo matching theory is deeply discussed. Additionally graphical theory which combines structural information of objects in stereo matching is described to make a theory foundation of related researches.Secondly, reference to the impersonal rules of human visional perceiving system, this thesis designs a course-to-fine frame of building extraction, which contains two steps: building location and extraction. In this thesis discusses a building location algorithm which combines multi-scale analysis theory and graph theory. In this algorithm, multi-scale characteristic of objects are expressed and these point features are properly described and buildings are located in large remote sensing images. Furthermore, in building extraction section, focus on the projective aberrance and occlusion problem, we study a precise building extraction theory based on inducted building models, in which projective aberrance and occlusion problem are solved by the optimizing process combined projective transformation parameters of object scene.Thirdly, aiming to sparse stereo matching of building features, this thesis discuss a sparse point stereo matching algorithm based on regional weighted adjacent graph, which will solve three problems in sparse stereo matching. In this algorithm, it utilizes spatial constrains to conquer the matching ambiguity problem, utilizes structural characteristic of point set to solve projective aberrance problems and utilizes support relation among point set to conquer image brightness and contrast changes due to various causation.Finally, this thesis discusses dense a stereo matching algorithms to cover the whole object scene based on sparse matching results. Focus on the points under occlusion and rupture, this algorithm utilizes visional psychology to choose adaptive support weight as the matching measure rules, which can sufficiently detected points under occlusion and rupture. Based on ASW rules, we discuss a quasi-dense stereo matching algorithm using SIFT and ASW rules. Furthermore, proper matching relation spreading strategy is discussed and we research a dense stereo matching algorithm combined spatial constrain and regional constrain to complete the dense stereo matching covered the interest object region. |
