Space Target Pose Estimation Based on High-Level Geometric Correspondences for Single-Photon LiDAR

Single-photon LiDAR is considered to be one of the most promising techniques for space target pose estimation due to its low power consumption, small size, and lightweight. And it is widely used in space target rendezvous, docking, and landing. However, such processing needs to address plenty of high-noise photon point clouds with limited resources and be completed in a very short time. To address these challenges, a global optimization method of pose estimation based on high-level geometric correspondences is proposed in this work. First, a K-Nearest Neighbors undirected graph is designed for describing the geometrical consistency of correspondences and a degree matrix is calculated to remove the false correspondences. In addition, a robust error metric based on Cauchy's function is introduced, which can effectively speed up the convergence of the optimization and iteratively update the parameters of the transformation matrix. The process of scanning a satellite with the single-photon LiDAR to obtain photon point clouds was simulated using the Monte Carlo method. Simulation results show that the proposed method can accurately estimate the pose of the satellite with a noise rate of 5%, 10%, and 15% in a time less than 1 second, revealing that it is of an obvious advantage in both efficiency and robustness when dealing with large-scale and noisy photon point clouds.