A binocular location method based on feature point matching and region pixel matching

Abstract

Aiming at the problem of welding failure due to insufficient positioning accuracy in binocular positioning of welding robot, an improved algorithm based on traditional binocular image stereo matching algorithm Semi-Global Block Matching (SGBM) is proposed in this paper. Firstly, the feature points of the whole image are matched through Scale-invariant feature transform (SIFT) corner detection algorithm, then the SGBM algorithm is used for regional stereo matching, the results of sift corner detection are integrated into SGBM matching algorithm. Finally, the parallax image obtained by matching is transformed into three-dimensional coordinates, so as to realize the welding target positioning of welding robot. Experimental results show that the accuracy of the improved algorithm is improved.