Research On Weld Tracking Algorithm Based On Correlation Filter

With the development of machine welding technology,robot automatic welding technology has basically replaced manual welding in actual production.The traditional teaching welding robot,on the one hand,takes a long time,and on the other hand is sensitive to the irregularity and discontinuity of the weld surface,so the efficiency is low.Therefore,it is of great significance to develop a welding system that can automatically identify and accurately track welds.Combined with the development status of welding seam detection technology and automatic welding seam tracking technology at home and abroad,as well as the difficulties existing in the current welding process,this thesis takes different types of welds as the research objects,and detects the initial feature points of the welds and analyzes them based on correlation filtering.The welding seam tracking algorithm is studied,and the main contents are as follows:（1）Research on feature point detection algorithm for different types of weld images obtained by vision sensors.The image preprocessing operation is performed on the collected weld image,and the structured light extraction algorithm based on the gray-scale centroid method is used to extract the fringe centerline in the image.According to the fringe centerlines extracted from images of different weld types,different feature point extraction algorithms are designed.The method of minimum intercept and line segment fitting is used to extract the weld feature points of the V-shaped butt weld image,and the second-order difference method is used to extract the weld feature points of the I-type butt weld and lap weld images.（2）Research the welding seam target tracking method based on correlation filtering algorithm.Using the feature point target area obtained by the welding seam feature point detection algorithm,the target area is used as a positive sample and a series of negative samples obtained by cyclic sampling to jointly train the correlation filter model.The response value is obtained to obtain the welding seam feature point area in the current frame image,so as to realize the welding seam target tracking.Based on the KCF（Kernelized Correlation Filtering）algorithm,the weld target model is established,and the feasibility of the algorithm is experimentally verified in various types of weld data sets.The experimental results show that the target tracking algorithm using correlation filtering can realize the tracking of the target of the weld.（3）In view of the scale change and occlusion problems in the KCF weld tracking process,improvements are made on the basis of the kernel correlation filter tracking algorithm.Combining the DSST（Discriminative Scale Space Tracker）scale adaptive algorithm and the TLD（Tracking,Learning and Detection）anti-occlusion algorithm,an anti-occlusion scale adaptive seam tracking algorithm is designed.The adaptive correlation filter template and occlusion judgment update mechanism are added to track the weld stably,and the method of combining tracking and detection is used to reposition the weld when the target feature point of the weld disappears.The algorithm and the benchmark algorithm are used to conduct welding seam tracking comparison experiments in various weld seam data sets.The experimental results show that the improved algorithm has better performance in the weld seam tracking data set than the benchmark algorithm,and improves the tracking accuracy and stability.Finally,through the on-site welding seam tracking experiment on the welding robot system platform,it is proved that the algorithm in this paper can better meet the welding seam tracking requirements under normal welding conditions.