一种图像立体匹配的误匹配点剔除与校正方法

A Method to Eliminate and Correct False Matching Point For Stereo Vision

针对双目立体视觉匹配中,图像的低纹理区域及重复纹理区域容易出现误匹配的问题,基于视差连续性及三角剖分,提出了一种误匹配点剔除与校正方法。首先利用SURF算法进行特征点提取,采用双向匹配策略进行特征点匹配;然后基于视差约束条件剔除误匹配点,并用三角剖分对误匹配点进行校正。实验结果表明,该方法能够剔除误匹配点,为误匹配点找到正确匹配点,有效提高双目立体匹配的精度。

There is usually some mismatching if the images present low texture area and repeated texture area in stereo vision system. In order to reduce mismatching, one way which can choose and correct the false matching points is proposed based on disparity continuity and Delaunay triangulation. The rough matching result is achieved by the SURF algorithm and the strategy of bidirectional matching. Then the disparity continuity constraint is used to eliminate the error matching. Moreover, the true matching feature points of the false matching points picked out can be attained by Delaunay triangulation. Experiments demonstrated that the method could effectively improve the accuracy of binocular stereo matching.