Real-time laser stripe robust detection for robotic weld seam tracking based on temporal image sequences

Vision-based seam tracking is fundamental to advancing the automation of robotic welding systems. To enable robust, real-time performance, we propose what is believed to be a novel lightweight segmentation algorithm. By leveraging temporal continuity, the method achieves superior immunity to noise like arc flash. Additionally, to enhance inference efficiency on edge devices, a co-optimization framework is designed to perform efficient, automated structured pruning of the segmentation model. Experiments confirm that the system operates at 55.2 FPS with a 0.154 mm tracking error, demonstrating its industrial efficacy.