宽基线立体影像Harris-Laplace特征的最小二乘匹配算法

Wide baseline image sequence least squares matching algorithm of Harris-Laplace

针对宽基线立体影像匹配的困难,本文提出一种基于Harris-Laplace的最小二乘匹配算法。算法基于Harris-Laplace特征检测器获得精度较高的初始特征点,对初始特征点进行NCC匹配,并应用基本矩阵F与单应矩阵H估计剔除误匹配点对,采用距离加权最小二乘匹配算法进行扩展匹配并同时保留定位精度较高的原始Har-ris-Laplace特征点。实验表明,此算法匹配率高,在视点改变、光照条件变化等情况下具有较好的鲁棒性。

An Harris-Laplace least squares matching algorithm was proposed for wide-baseline stereo image matching in the pa- per. In this algorithm, high accuracy of initial matching points was obtained with Harris-Laplace detector, other Harris-Laplace features were matched by using distance-weighted least squares matching algorithm while retained the original precise features of Harris- Laplace points. Experiments demonstrated that the algorithm was good at matching, and this method had good robustness when viewpoint varies and lighting conditions changes, etc.