A novel laser stripe key point tracker based on self-supervised learning and improved KCF for robotic welding seam tracking

Laser vision based real-time welding seam tracking has emerged as a potent strategy for enabling intelligent robotic welding. And trackers based seam key point tracking algorithms demonstrate remarkable adaptability to complex welding environments. This paper proposed a self-supervised robust KCF (Kernelized Correlation Filter) tracker for seam key point tracking, which could be a novel approach to achieve autonomous seam tracking. Firstly, a self-supervised global-local feature extraction network is constructed, which can guide the model to focus on both global semantic and local texture features of laser stripes, thereby establishing a solid groundwork for stable key point tracking. Subsequently, a robust KCF tracking algorithm is presented. A multi-template enhanced tracker generation strategy is designed, and the corresponding analytical solution is derived, which can improve the tracker's representation capability of stripe features without significantly increasing computational complexity. Experimental results demonstrate that compared to traditional algorithms, the proposed algorithm exhibits advantages in tracking accuracy, stability, and real-time performance. Moreover, since the algorithm minimally relies on manually labeled data, it holds promise as a technological means to achieve fully autonomous seam tracking in actual welding production.