A multi-BRIEF-descriptor stereo matching algorithm for binocular visual sensing of fillet welds with indistinct features

Abstract

Binocular vision is employed in the present work for recognition and positioning of weld seams with the expectation of well handling the reconstruction of the welds for extracting the weld information. However, the indistinct or similar features for the surface of weldments make the stereo matching process in binocular vision difficult. A multi-BRIEF-descriptor stereo matching algorithm with the combination of epipolar constraint is proposed to improve the matching results and to reduce the time consumption. In the algorithm, the distinct edge information of the weldment is carefully detected and taken as the feature points, and the multiple stereo matching processes are conducted with different BRIEF (Binary Robust Independent Elementary Features) descriptors to eliminate the mismatches caused by the indistinct or similar features with fused information. Stereo matching results show that this approach has advantages in reducing the mismatches and obtaining more matched feature points compared with the feature matching methods with the descriptor of SURF (Speeded Up Robust Features), or ORB (Oriented FAST and Rotated BRIEF), although it consumes somewhat more time than that with SURF.