Robust welding seam tracking and recognition

In the process of automatic welding based on structured light vision, the precise localization of the welding seam in an image has an important influence on the quality of the welding. However, in practice, there is much interference, such as spatter and arc, which introduces great challenges for accurate welding seam localization. In this paper, we considered welding seam localization problem as visual target tracking and based on that, we proposed a robust welding seam tracking algorithm. Prior to the start of welding, the seam is separated using a cumulative gray frequency, which is utilized to adaptively determine the initial position and size of the search window. During the welding process, large seam motion range may result in only a portion of the welding seam exists in the search window. To prevent that, a tracking-by-detection method is used to calculate the location of the search window. Usually, the intersection of seam and noise, e.g., spatter, has a severe influence on the accuracy of feature points localization. In order to solve this problem, a sequence gravity method (SGM) for extracting a smoother center line of welding seam is proposed, which is able to reduce the impact of interference. The double-threshold recursive least square method is used to fit the curve obtained by SGM with the aim of improving the real-time performance and accuracy of the system. Finally, the superiority of the proposed algorithm is well demonstrated by comparison with other solutions for seam tracking and recognition through extensive experiments.