Research on registration technology of steel structure parts

Steel structure parts are widely used, and welding robots have improved their welding efficiency. However, 3D cameras fail to collect complete point cloud data before automatic welding, causing insufficient weld track extraction and affecting quality. This paper proposes a registration method combining the 4 - Point Congruent Sets (4PCS) for coarse registration and improved the Iterative Closest Point (ICP) for fine registration. 4PCS initializes model and actual point cloud alignment. Improved ICP uses the k - dimensional tree (kd-tree) for rapid nearest-neighbor search and normal vector constraints to eliminate wrong point pairs. Experiments show average ICP error is 2.91mm, reduced to 2.61mm by improved ICP, proving higher accuracy.This research contributes to the technical field of robotic welding automation, specifically addressing the quality control challenges caused by incomplete point cloud data in steel structure welding. It provides a reliable registration method for weld seam extraction systems, enhancing the stability of robotic welding operations.