一种基于 Harris-Sift 引导 LK 光流约束的特征点匹配算法

摘要

A image matching algorithm based on Harris-Sift guides Lucas-Kanade optical flow is presented. Firstly, the Harris-Sift feature points are extracted on the images and matched roughly by bidirectional nearest neighbor matching algorithm. Then, the two-views homography matrix is robustly estimated through Ransac algorithm, and then, the matrix is used to guide Lucas-Kanade optical flow to search locally optimal matching points for getting many more quantity and precise matching points. Finally, the outer points are removed by the epipolar constraint. Experimental results show that the algorithm can decrease the probability of the outer points while improving the quantity of the matching points effectively, and the computing time can be saved.%提出了一种基于 Harris-Sift 特征点引导 LK 光流约束的图像匹配算法。算法首先检测图像的 Harris-Sift 特征点，利用双向互匹配的最邻近搜索算法进行粗匹配；然后利用 Ransac 算法鲁棒估计两视图单应矩阵，利用单应矩阵引导 LK 光流法寻找局部最优匹配点，以获取更多更精确的匹配点；最后利用极线约束剔除外点。实验结果表明，该算法能够在降低外点概率的同时有效提高匹配点的数量，并能节省运算时间。