基于稀疏光流法的ORB特征匹配优化

针对图像特征误匹配数量大的问题,提出一种基于稀疏光流法的ORB图像特征点匹配算法。对特征点进行暴力匹配得到初始匹配点集,利用稀疏光流法计算特征点运动向量,估计出特征点在待匹配图像中的二维坐标位置,剔除偏离估计位置较远的特征点匹配对,最后利用随机抽样一致算法进行几何校验进一步优化匹配结果,达到剔除误匹配的效果。实验结果表明:该算法相较于ORB算子、SIFT算子、SURF算子准确率平均提升了21.6%,较RANSAC-ORB算法准确率平均提升了2%,且该算法对图像光照变换、视角变换、模糊变换、旋转和缩放变换和光照变化具有较好的通用性。

ORB feature matching optimization based on sparse optical flow method

Aiming at the problem of large number of image feature mismatches, an oriented FAST and rotated BRIEF (ORB) image feature point matching algorithm based on sparse optical flow method was proposed. Firstly, the feature points were violently matched to obtain the initial matching point set. Then the sparse optical flow method was used to calculate the feature point motion vector, and the two-dimensional coordinate position of the feature point in the image to be matched was estimated, and the feature points far from the estimated position were eliminated. Finally, the random sampling consistency algorithm was used to perform geometric verification to further optimize the matching result, so as to eliminate the effect of mismatching. The experimental results show that compared with the ORB operator, SIFT operator and SURF operator, the accuracy of this algorithm is increased by 21.6% on average, and the accuracy of random sample consensus FAST and rotated BRIEF (RANSAC-ORB) algorithm is increased by 2% on average; moreover, the algorithm has good versatility for image illumination transformation, perspective transformation, fuzzy transformation, rotation and scaling transformations and illumination variations.