Application of Micro-Plane Projection Moving Least Squares and Joint Iterative Closest Point Algorithms in Spacecraft Pose Estimation

Abstract

Accurately determining the attitude of non-cooperative spacecraft in on-orbit servicing (OOS) has posed a challenge in recent years. In point cloud-based spatial non-cooperative target attitude estimation schemes, high-precision point clouds, which are more robust to noise, can offer more accurate data input for three-dimensional registration. To enhance registration accuracy, we propose a noise filtering method based on moving least squares microplane projection (mpp-MLS). This method retains salient target feature points while eliminating redundant points, thereby enhancing registration accuracy. Higher accuracy in point clouds enables a more precise estimation of spatial target attitudes. For coarse registration, we employed the Random Sampling Consistency (RANSAC) algorithm to enhance accuracy and alleviate the adverse effects of point cloud mismatches. For fine registration, the J-ICP algorithm was utilized to estimate pose transformations and minimize spacecraft cumulative pose estimation errors during movement transformations. Semi-physical experimental results indicate that the proposed attitude parameter measurement method outperformed the classic ICP registration method. It yielded maximum translation and rotation errors of less than 1.57 mm and 0.071°, respectively, and reduced maximum translation and rotation errors by 56% and 65%, respectively, thereby significantly enhancing the attitude estimation accuracy of non-cooperative targets.

Keywords

• on-orbit service
• ToF camera
• point cloud
• pose estimation
• moving least squares