A robust algorithm for surface corner matching based on homograph

Summary form only given. Stereo matching is an important research area in stereovision. This paper uses reprojection error to compute the homography matrix to match the feature points. First original matching is determined by epipolar constraint. The image pair is partitioned to a serial of planes under the minimal reprojection error. Delaunay triangulation method is used and the original matched point pairs act as vertexes. In particular, if the scene contains planes line segmentation is used to group the corners. Then we check coplanar points in the 3 dimensional positions to compute the homography matrix of each compositive plane. We search the corresponding point of the plane near the corresponding epipolar line. The method can resolve the occluded problem in stereo matching effectively.