Optimization-Based Geometric Enhancement and Motion Estimation for Non-Cooperative Spacecrafts

Abstract

The state estimation of non-cooperative spacecrafts is a crucial prerequisite for on-orbit services. Aiming at the challenges in the fusion-based scheme with monocular vision and sparse point cloud, an optimization-based method of geometric enhancement and motion estimation is proposed in this paper. First, with the novel idea of geometric shape representation using simple features, a real-time segmentation framework is established. Differing from segmentation models, it can guarantee both complete segmentation and high inference speed. Second, given the assumption of local shared planes, a new label-free algorithm of point cloud densification is developed with an explainable model. To improve its efficiency, a curvature-guided strategy is employed to sample depth-incomplete points conducive to feature enhancement. Compared with sparse point clouds, it shows higher pose observation accuracy. Third, a truncation compensator is built to fit the high-order terms of a nonlinear state transition model with online optimization, which mitigates the impairment in a priori estimation. Combined with the adaptive extended Kalman filter, the motion can be estimated with fewer errors. Finally, the proposed method is validated through comparative simulations and ground experiments.