一种基于区域立体匹配算法的实现及改进

在立体视觉中，立体匹配是立体视觉中算法中最重要也是最难的部分。如何更快更精确的建立两幅图像的匹配基元之间的对应关系，以及解决区域匹配算法计算量大等缺点，一直是立体匹配中的热点。文中首先将图象进行外极线校正，在此基础上，采用一种隐含约束条件结合唯一性约束条件的单向匹配算法(SMP算法)，完成了初次匹配及初次去除伪匹配，再采用一种伪极线约束条件，对得到的匹配对进行二次去除伪匹配，最后，对得到的视差图进行插值运算，提高了匹配精度。另外，在匹配过程中，通过Box滤波加速方法，减少了计算匹配窗口相似性的复杂度，缩短了程序运行时间，可以满足系统对实时性的要求。

The implement and improving of one area-based stereo matching algorithm

Stereo matching is the most important and difficult part in the stereo vision. People have been working on how to establish matching cells between two homologous images more accurately and decrease the computational cost during matching. In the paper, firstly the images are rectified which makes the search range from two dimension to one dimension,and then adopt a latent constraint and one algorithm named SMP algorithm which is based on uniqueness constraint to achieve the matching of stereo pairs. This paper proposse one new method which make SMP algorithm more accurate than before.Using false constraint is to detect some fake match points gotten by SMP algorithm ,and get disparity map after interpolation.In addition, implementing Box filtering accelerating method is to decrease the computational cost and reduce program running time which is satisfied with application of real-time system.