A Registration and Fusion Method of 3D Cross-source Point Cloud Data for Modeling Accurate Models of Small Celestial Bodies

High-precision three-dimensional models of small celestial bodies are important for deep space exploration tasks. The homologous point cloud obtained from a single sensor has limitations in terms of accuracy and density. For example, the point cloud obtained by RGB camera has rich information and high density, but low precision. The point cloud obtained by laser scanning has high precision but low density. Therefore, we hope to register and fuse the point clouds obtained from different sensors, synthesize the advantages of cross-source point clouds obtained from different source domains, and improve the density, accuracy and other aspects of small celestial point clouds. To solve the common problems of cross-source point clouds, we adopt an end-to-end feature measurement point cloud registration framework, which does not need to search the corresponding points between two point clouds, and reduces the difficulty of processing cross-source point cloud data. Aiming at the problem that some stray points on the registered point cloud surface affect the accuracy of the point cloud, a point cloud fusion optimization method based on Euclidean distance is proposed. The stray points around the point cloud are removed to reduce the loss of normal points. The accuracy of point cloud is improved by the fusion after registration. Experiments show that the proposed method can effectively improve the density and accuracy of small celestial body models.