| Image-based modeling and rendering is currently one of the most challenging topics in Computer Vision and Photogrammetry. Given two images of the same scene, the goal is to build a dense set of point pairs, one from each image, that represent the same point feature in the real scene; this is called dense matching or stereo matching. Among all dense matching algorithms, Semi-Global Matching (SGM) is considered to be one of the most promising algorithms for real-time stereo vision. SGM does not suffer from the classical "streaking problem" of other approaches and has greatly improved accuracy and efficiency. The aim of this thesis is to further improve the accuracy of SGM without increasing the computational cost. We propose setting the penalty parameters adaptively according to image edges extracted by edge detectors. When applied to the standard Middlebury stereo dataset, we have achieved noticeable accuracy improvements without increased computation time. |
