An Adaptive Point Cloud Registration Method for Weld Beads Based on Geman-McClure Robust Kernel Function

With the purpose of satisfying the measurement requirements for large-scale weld beads in industrial welding components, it is essential to reach precise stitching of multiple acquired point cloud datasets for achieving accurate 3D reconstruction. The registration of overlapping point cloud regions constitutes a critical step in this stitching process, where the accuracy of the registration method significantly impacts the final reconstruction quality. Therefore, an adaptive point cloud registration method was proposed in our study based on the Geman-McClure (GM) robust kernel function. First, the GM robust kernel function was employed as the error metric in the ICP algorithm to effectively suppress the interference of outliers on overall registration errors while significantly reinforcing algorithmic accuracy and stability. Second, a MAD-based adaptive scale adjustment mechanism was introduced to dynamically regulate scale parameters following the distribution characteristics of the point cloud, contributing to the strengthened adaptability of the algorithm to complex data distributions. Furthermore, experimental validation with public datasets reveals that the proposed algorithm maintained high registration accuracy when processing outlier data. In the weld bead registration experiments, the point cloud data was initially preprocessed through co-ordinate system unification, and subsequently, the proposed algorithm was employed to perform fine registration of overlapping regions. The results suggest that the pro-posed algorithm achieved high accuracy in real-world weld bead point cloud registration scenarios.