An Improved GMS Image Feature Matching Algorithm Based on BEBLID Descriptor

Aiming at the shortage of matching accuracy of ORB algorithm in image matching stage and the poor robustness of GMS algorithm under repeated texture conditions, this paper proposed an improved GMS image feature matching algorithm based on BEBLID descriptor. This algorithm firstly uses the ORB algorithm for image feature point extraction, secondly describes the feature points with BEBLID descriptor, after that discards false matching pair preliminarily by brute force matching. In order to improve the matching accuracy of the algorithm, two algorithms GMS and PROSAC are combined on this basis to obtain better matching pairs. The experimental results show that the algorithm has uniform extraction of feature points and high matching accuracy for different image feature matching, and its correct rate is improved by 10.97 percentage points to the GMS algorithm, which can meet the demand of large parallax image matching and improve the accuracy and efficiency of target acquisition in vision tasks.