基于改进 SIFT 和多约束的 UAV 影像匹配方法

针对尺度不变特征转换(scale invariant feature transform, SIFT)算法在无人机(unmanned aerial vehicle, UAV)影像的匹配过程中存在特征点稳定性差和误匹配多的问题,文章提出一种基于改进SIFT和多约束的UAV影像匹配方法。首先,在对影像降采样后,综合采用SIFT算法和Scharr-ORB(oriented brief)算法共同进行特征点检测和描述;然后,使用最近邻距离比值法(nearest neighbor distance ratio, NNDR)、双向约束匹配和余弦相似度约束匹配的多约束方法进行特征点的粗匹配;最后,使用最小中值(least median of squares, LMedS)算法计算基础矩阵和随机抽样一致性(random sample consensus, RANSAC)算法计算单应矩阵的多约束方法进行特征点的精匹配,进一步提高匹配精度。结果表明,该方法在获得更多特征点和匹配对数的同时,能够剔除较多的误匹配,使其具有较高的匹配正确率和匹配精度。

Aiming at the problems of instability of feature points and many mismatches in the process of unmanned aerial vehicle (UAV) image matching with scale invariant feature transform (SIFT) algorithm, a UAV image matching method based on improved SIFT and multiple constraints is proposed. Firstly, after the image downsampling, SIFT algorithm and Scharr-ORB algorithm are employed to detect and describe the feature points. Then, a rough matching of the feature points is established through multi-constraint methods of nearest neighbor distance ratio (NNDR), bidirectional constraint matching, and cosine similarity constraint matching. Finally, a multi-constraint method using least median of squares (LMedS) algorithm for fundamental matrix computation and random sample consensus (RANSAC) algorithm for homography matrix computation is adopted for the fine matching of feature points to further improve the matching accuracy. Experimental results show that this method can get more feature points and matching pairs while eliminating more mismatches, which makes the algorithm have higher correct matching rate (CMR) and matching accuracy.