Abstract: Image matching is the main content of computer vision application research.Aiming at the problems of Oriented fast and Rotated BRIEF (ORB) image matching method with no scale invariance and low matching accuracy,an image matching method based on improved ORB algorithm was proposed.In the feature point detection stage,the feature points were detected by ORB and SURF at the same time.The left and right image feature points were detected by oFAST and SURF algorithm,and then the feature points were described by rBRIEF descriptor.In the stereo matching stage,based on the rough matching of Hamming distance on the feature points,the epipolar constraint was introduced to screen the feature points and make fine matching,so as to reduce the matching search range and speed up the matching speed to improve the matching accuracy.The experimental results showed that the average point logarithm of the improved ORB algorithm was about 1.5 times of SURF algorithm,the average matching speed was 22% higher than SURF algorithm,the accuracy rate was 2 times higher than SURF algorithm and 1.7 times higher than ORB algorithm.In conclusion,the improved ORB algorithm had the characteristics of more matching points,faster speed,higher accuracy,and scale invariance.It could be used in target recognition,target tracking,3D reconstruction,defect detection and other fields.

Key words: improved oriented FAST and rotated BRIEF algorithm, feature point detection, feature point description, epipolar constraint, image matching

摘要： 图像匹配是计算机视觉应用研究的主要内容。针对ORB(Oriented FAST and Rotated BRIEF)图像匹配方法不具备尺度不变性和匹配精度低的问题,提出一种基于改进ORB算法的图像匹配方法。在特征点检测阶段,ORB和SURF(Speeded Up Robust Feature)同时检测特征点,首先采用oFAST(oriented-FAST)与SURF算法检测左右图像特征点,然后使用rBRIEF(rotation-aware BRIEF)描述子描述特征点;在立体匹配阶段,采用Hamming距离对特征点进行粗匹配的基础上,引入极线约束筛选特征点并进行精匹配,减小匹配搜索范围,加快匹配速度,提高匹配准确率。实验结果表明,所提改进ORB算法点对数平均值是SURF算法的15倍左右,匹配速度平均值比SURF算法提高了22%,准确率比SURF算法提高了2倍,比ORB算法提高1.7倍,从而证明所提改进ORB算法,具有匹配点数多、速度快、准确率高的特点,且具有尺度不变性。所提方法可应用于目标识别、目标跟踪、三维重建、缺陷检测等领域。

关键词: 改进ORB算法, 特征点检测, 特征点描述, 外极线约束, 图像匹配