Cross-Source Point Cloud Registration Algorithm Based on Multiple Filters

With the rapid development of sensing technologies, cross-source point clouds are more convenient and widely used in practical point cloud registration compared to same-source point clouds. However, the registration of cross-source point clouds is more challenging due to density differences, scale variations, and data missing issues. Addressing the challenges of cross-source point clouds, this paper proposes a cross-source point cloud registration algorithm based on multiple filters. In the preprocessing stage, the algorithm utilizes multiple filters to denoise and down-sample the point cloud data, effectively addressing the density differences in cross-source point clouds. Subsequently, point feature histograms (FPFH) are computed to obtain feature point pairs, and a scaling factor is introduced to initially estimate the scale differences between the two sets of point clouds. In the registration phase, a coarse registration is performed using the SAC-IA algorithm, followed by fine registration using a multi-scale adaptive ICP algorithm. To validate the effectiveness of the algorithm, human back point clouds are scanned using a laser scanner and a Realsense D455 depth camera. Comparative experiments with other algorithms of similar type are conducted. The results demonstrate that, in cross-source point cloud registration, the proposed method outperforms other point cloud registration methods, showing superior performance.