Weld feature extraction and location based on corner detection and optical flow tracking
                                         LIN Shaoduo1, GAO Xiangdong1, LI Yangjin1, ZHANG Nanfeng1, QUAN Fanghong2
(1.Guangdong Provincial Welding Engineering Technology Research Center, Guangdong University of Technology, Guangzhou 510006, China; 2.Guangdong
 Forging Machine Tool Factory, Foshan 528300, China)

Abstract: Aiming at the detection of weld feature points in Vshaped butt welding of groove plate, a fast extraction and location method of weld feature based on laser vision sensor corner detection and optical flow (lucas kanade, LK) tracking was studied. The laser vision sensor which can detect the weld seam feature image in real time was designed according to the principle of triangulation, and a mathematical model from the pixel coordinates of the laser stripe feature points to the 3D coordinates of the weld feature points was established. Weld seam images were preprocessed and the weld features were extracted by using ShiTomasi corner detection. Finally, the subpixel position of the weld seam feature points in the image was calculated in real time by using the optical flow method to match the feature corners for the subsequent frames. Experimental results indicate that the proposed method of feature extraction and location based on corner detection and optical flow tracking has a higher detection accuracy and the average error is less than ±0.13 mm, which can identify the weld features in real time and accurately.

Key words: corner detection; optical flow tracking; weld feature location; laser vision