一种基于极几何和单应约束的图像匹配算法

Algorithm for Images Matching Based on Epipolar and Homography Constraints

提出一种综合应用极几何和单应约束的图像特征点匹配算法,首先使用互相关法对图像特征点集进行初始匹配,然后运用RANSAC方法鲁棒地估计基本矩阵和单应矩阵并相应地剔除错误匹配点,最后利用优化后的基本矩阵和单应矩阵引导匹配以获得更多、更精确的匹配点。大量真实图像实验表明,所提出的算法能够产生更多的匹配点并具有较高的匹配精度。

An algorithm for images matching was proposed, which used both epipolar and homography constraints. First, the cross-correlation matching was used to get the initial matching, then the fundamental matrix and the homography matrix ware estimated robustly with RANSAC algorithm, at the same time, most of wrong points ware picked and deleted. At last, using optimized fundamental matrix and homography matrix, more precise matching points could be obtained through the guide matching. Lots of experiments show that this algorithm can bring more matching points also with high precision.