Disparity-Space Images and Large Occlusion Stereo

摘要: A new method for solving the stereo matching problem in presence of large occlusion is presented. data structure — disparity space image defined which we explicitly model effects regions on solution. We develop a dynamic programming algorithm that finds matches and occlusions simultaneously. show while some cost must be assigned to unmatched pixels, our algorithm's occlusion-cost sensitivity algorithmic complexity can significantly reduced when highly-reliable matches, or ground control points, are incorporated into process. The use points eliminates both need biasing process towards smooth solution task selecting critical prior probabilities describing formation.