Research On Stereo Matching Algorithm Based On Image Line Segments

Stereo matching is the most difficult issue in stereo vision. The target of stereo matching is finding the correspondence features from several images and calculating the disparity value of the every feature in the reference image.In this paper, the problem of stereo matching is divided into the choice of matching element and the rule of matching. The stereo algorithm which has four key components including feature space, similarity measure, search space and search strategy, is deeply analyzed based on a full and system research of stereo algorithm.This paper makes how to improve the accuracy and the speed of stereo matching as the research topic. A global algorithm is presented, which can be divided into two steps: the extraction of line segment and tree dynamic programming.First, self-adaptive threshold is applied to detect the canny edge of the reference image, which makes full use of accurate positioning of edge detection and the high effectivity of image segmentation. Each scanline of the reference image is divided into several segments. The accuracy and the speed of algorithm can be guaranteed because the length of segment can be adapted.Then, a spanning tree (ST), which is the most effective global stereo matching algorithm, is constructed with segments as the nodes of the tree. This paper applies dynamic programming to a tree structure, as opposed to the individual scanline, which adds the constraints between different scanlines. Thus our algorithm is a global optimization method because disparity estimate at one line segment depends on the others. Experiment results demonstrate that this algorithm can meet both demands for high accuracy and real time.