Pose estimation of space targets based on keypoint completion and weighted least squares optimization

Accurate pose estimation of space targets are critical for autonomous spacecraft navigation and on-orbit operations. Visual cameras offer rich image data for accurate pose estimation. However, changes in lighting and occlusion often cause missing or incorrect keypoints, thus reducing the robustness of the method. To address this issue, this paper proposes a novel method called Keypoint Completion and Weighted Least Squares Optimization (KC-WLS). The method first employs a Pose-Guided Keypoint Completion (PGKC) module that uses the historical pose of target to infer and recover missing keypoints. Subsequently, a Dynamic Weighted Least Squares (DWLS) module is used to optimize the pose estimation. The dynamic weighting strategy adaptively balances the contributions of detected and recovered keypoints. Experimental results show that the proposed method achieves 97.86 % pose estimation accuracy for typical space targets within a 1 km range when all keypoints are detected. Moreover, it maintains 91.44 % accuracy even with only three detected keypoints. These results highlight the robustness and adaptability of our method, offering a reliable solution for autonomous space perception and navigation. • Pose-Guided Keypoint Completion recovers missing keypoints using historical pose data. • Dynamic Weighted Least Squares balances keypoint contributions via dynamic weighting. • We achieves 91.44 % accuracy with 3 keypoints and 97.86 % with all keypoints.