A New Stereo Matching Method Based on Sub-pixel Corner Detection

Corner detection and stereo matching play an important role in the machine vision. The detection accuracy and matching velocity are highly demanded in the stereo vision. In this paper, dual-Harris corner detection algorithm and corner matching method are proposed. Corner detection algorithm adopts random mid-point displacement method in fractal geometry to subdivide and amplify the pixel precision corner detected by the Harris algorithm and its surrounding pixel block. And then, it reuses harris algorithm to provide sub-pixel corner position by detecting the fractal interpolation map and the FBS surface. This method accords with human thought manner that is to observe things ranging from whole to part and form macroscopy to microcosm, which avoids detecting the edge and fitting lines by comparing with the common sub-pixel detection algorithm. Corner matching method uses the corner characteristic constraint to limit the amount of the alternative corners in the searching window, which sharply decreases the amount of correlation coefficient computation, and saves the stereo matching time. Simultaneity, the constraint of disparity gradient is selectively adopted to increase the stereo matching veracity. Large numbers of experiment results show the veracity and real-time of the proposed method.