基于渐进可靠点生长的散斑图快速立体匹配

Fast Stereo Matching Based on Progressive Reliable Point Growing Matching for Speckle Pattern Images

传统的区域生长匹配结果过于依赖生长阈值,而渐进可靠点生长匹配设置由高到低的种子点选取阈值和生长阈值迭代地生长区域,有效增强了匹配结果对于生长阈值的鲁棒性。有研究将该方法用于散斑图的立体匹配,取得了较好的效果,但是计算速度慢且仍存在不少非法点。实现了一种划分格网进而筛选种子点的方法,并在生长过程中丢弃传统的四邻域生长,提出一种跳跃式生长方法,其在确保匹配质量的同时大大提高了计算速度。进一步地,将渐进可靠点生长的方法用在匹配后处理中消除了一半的非法点。

Different from the traditional region growing matching algorithm which is excessively dependent on the growing threshold,PRPGM(Progressive Reliable Points Growing Matching)sets a sequence of thresholds from high to low for both seed point selection and growing matching to implement the stereo matching in an iterative process,thus enhancing the results' robustness on the thresholds.A research has employed this method for depth estimation for speckle projection system and achieved good results.However,it motivates us to conduct this research that the calculation is not fast enough for practical applications as well as there are still many illegal points on the disparity map.We proposed a sparse seed point selecting method that selects seed point from grid divided by certain size on the speckle pattern image in the stage of selecting seed point and a jumping growing matching method instead of the conventional connectivity constraint of 4-neighbors domain during the growth of the reliable points.Both of the above revises have been verified to improve calculation speed greatly.Meanwhile,we applied the PRPGM to handle illegal points in postprocessing of stereo matching and eliminated most of illegal points consequently.