一种基于ORB的特征匹配算法

An ORB-based feature matching algorithm

SURF算法具有尺度不变性、旋转不变性和较好的鲁棒性但是不具备实时性,ORB算法良好的实时性却不具备尺度不变性,结合两种算法的优缺点提出了基于ORB和SURF的特征匹配算法(简称S-ORB)。第一步改进ORB算法提取特征的空间结构,集合SURF算法提取特征点;第二步构建ORB算法描述子;第三步进行特征匹配。在匹配的过程中,采用汉明距离完成初步匹配,然后结合RANSAC算法对初步筛选的特征点进行错误剔除,获取匹配较准确的特征点对。实验结果表明,图像尺度变化时,改进的算法在匹配准确度较SURF提升5倍,较ORB算法提升3倍,在特征点分布均匀性方面也有所改进。

The SURF algorithm has scale invariance,rotation invariance and good robustness but does not have real-time performance. The ORB algorithm has good real-time performance but does not have scale invariance. Combining the advantages and disadvantages of the two algorithms we propose a new algorithm based on ORB and SURF. The first step of feature matching algorithm(S-ORB)is to improve the spatial structure of feature extraction by ORB algorithm,and to extract feature points by SURF algorithm;the second step is to construct ORB algorithm descriptor;the third step is to perform feature matching in matching process. The Hamming distance is initially matched,and then the RANSAC algorithm is used to perform error elimination on the preliminary selected feature points to obtain a more accurate feature point pair. The experimental results show that the improved algorithm improves the matching accuracy by 5 times compared with SURF,3 times higher than ORB algorithm,and also improves the uniformity of feature point distribution.