A method for rebar weld seam extraction based on 3D point clouds

Abstract Currently, research on weld seam extraction mostly focuses on simple shapes, such as V-shaped or T-shaped seams. However, rebar weld seams exhibit complex characteristics. Existing rebar weld seam extraction methods often rely on simple geometric representations, such as line, plane, or cylindrical fitting, thus have limited capability in capturing complex spatial variations. This paper proposes a novel method for weld seam extraction from input point cloud. The proposed method does not impose explicit geometric assumptions on the weld seam shape. By utilizing principal direction consistency region growing and projection histogram segmentation, the proposed method is well suited for the seam identification and extraction of rebar structures. An experimental platform is set up for evaluating the effectiveness and robustness of the proposed method. Experiments show that the proposed method correctly extracts 29 weld seams out of 30 sets of testing point cloud, achieving a success rate of 96.7%, the precision, recall, and F1-score all exceed 0.9. Under different camera position offsets and distances, the average deviation in the ZY plane is all less than 1.5 mm. The experimental results demonstrate that the proposed method can accurately and robustly identify the weld seam in rebar structures, providing a solid data foundation for intelligent welding.