Multi-view vision fusion with coarse-to-fine detection for high-precision robotic workpiece pose recognition and welding guidance

To address robot challenges in autonomously recognizing workpiece poses and locating weld seams for large-scale components in unstructured environments, this study proposes a multi-view robotic system with coarse-to-fine detection. A composite sensor (line-structured light + monocular camera) mounted on the end-effector captures multi-view images. Hand-eye calibration unifies views into the base coordinate system, enabling point cloud reconstruction for pose calculation. Plane segmentation and geometric relations yield an initial seam pose, guiding high-precision structured light scanning along optimized trajectories. Real-world experiments on a large-scale workpiece demonstrate root mean square errors (RMSEs) of <7.92<i>m</i><i>m</i> in scanning path recognition and <0.613<i>m</i><i>m</i> in actual welding trajectories, meeting industrial requirements. Key innovations include a hybrid sensor for complementary detection, multi-view spatial fusion, and a hierarchical framework integrating geometric priors with data-driven scanning.