【摘要】 立体视觉技术被广泛的应用于视觉导航系统中.由于要求计算快速,以往的立体视觉导航系统通常采用局部匹配的方法,其匹配精确度较差,不能满足高性能视觉导航的需要.为此,我们提出了一种新的快速立体视觉匹配算法,该算法将每一条外极线分割成亮度一致的线段,并采用全局的树型动态规划进行优化求解,既充分利用了全图的相关信息,取得了较高的匹配准确度,又减少了搜索变量的个数,使得系统的效率得到大幅度的提高.实验结果表明,该算法能够取得与基于MRF的全局网格匹配算法相当的匹配结果,而执行时间上却很快,为视觉导航提供了可靠的环境三维数据.更多还原

【Abstract】 Stereo technology has been widely used in visual navigation systems.Traditional stereo vision navigation systems usually use local method because of the real-time requirement.The accuracy of the results by local method,however,is very low and it can not be used in high performance vision navigation systems.In this paper we propose a novel fast stereo matching algorithm,which utilizes the segmentation information on the epipolar lines to form a tree and get optimized results by using tree dynamic programming procedure.The proposed algorithm can both use much of the relation information between neighboring pixels to achieve a more accurate matching result,and reduce the number of vertices to be solved so that the system becomes very efficient.Experimental results have shown that our algorithm can get good results comparable to the global MRF-based algorithms,while the executing time is much shorter than the latter.Our algorithm is a fast algorithm which can provide reliable 3D information for vision navigation.更多还原