A Feature Extraction Approach Over Workpiece Point Clouds for Robotic Welding

The programming of welding trajectories hinders the development of welding robots. With the development of machine vision, the welding path planning based on 3D vision can eliminate the teaching and programming work of the robot. However, the current weld extraction research relies on the geometric information of the workpiece, resulting in the low generality and robustness of the existing methods. The edge and corner features of the point cloud are the basis for extracting welds. Therefore, in view of the above problems, this paper proposes a method for edge and corner extraction of workpieces applied to 3D vision welding. First, the LOBB feature descriptor is proposed to build a feature space for the workpiece point cloud. Then, to enhance the robustness of feature clustering, nonlinear activation is performed on the feature description. Finally, a hierarchical clustering method based on K-means is designed to achieve the extraction of crease points, boundary points, and corner points. Experiments show that the method in this paper can not only extract the edges of various workpieces completely and with low noise, but also can identify corners efficiently, which is overall better than the existing methods. Note to Practitioners —Welding trajectory programming has been a bottleneck in the development of welding robots. However, advancements in machine vision have opened up new possibilities for welding path planning using 3D vision, which can eliminate the need for manual teaching and programming of robots. Unfortunately, existing methods for weld extraction from point clouds rely heavily on geometric information, resulting in limited applicability and robustness. To address these challenges, this paper proposes a novel method for extracting edge and corner features from workpieces using 3D vision in welding applications. The key contribution is the introduction of the LOBB feature descriptor, which creates a feature space for the workpiece’s point cloud. By applying nonlinear activation to the feature description, the robustness of feature clustering is enhanced. Additionally, a hierarchical clustering method based on K-means is designed to extract crease points, boundary points, and corner points. Experimental results demonstrate that the proposed method can effectively extract edges from various workpieces with minimal noise and efficiently identify corners. Overall, the method outperforms existing approaches in terms of completeness and accuracy of edge extraction. This advancement holds significant potential for practitioners in the field of welding robotics, as it reduces the programming complexity and improves the reliability of welding robots in real-world applications.