Research On Multi-scale Adaptive Cross-source Point Cloud Registration Algorithm

With the rapid development of sensing technology and the demand for digital construction, three-dimensional data has been widely applied in many industries. In the processing and application of three-dimensional point clouds, point cloud registration is a crucial step, directly impacting the positioning accuracy and operational efficiency during application. This paper focuses on the research of a multi-scale adaptive cross-source point cloud registration algorithm, aiming to address the issues of density and scale differences in cross-source point cloud registration. Firstly, a combination of filters is employed in the preprocessing stage to denoise and down-sample the original point cloud, thereby addressing the density differences between cross-source point clouds. Additionally, differential down-sampling is applied to different parts of the same point cloud with varying densities, preserving more feature information in denser regions. Subsequently, the FPFH algorithm is improved to obtain matching point pairs, initially addressing the scale differences between cross-source point clouds. Finally, the SACIA algorithm is used for coarse registration of point clouds, followed by precise registration using the proposed multi-scale adaptive ICP algorithm. To analyze and validate the effectiveness of the algorithm, point clouds of the human body model’s back are obtained using both a laser scanner and a depth camera, and comparative experiments are conducted with several other registration algorithms. The results indicate that in cross-source point cloud registration, the method proposed in this paper outperforms other point cloud registration methods in terms of performance.