中国空间科学技术

中国空间科学技术 ›› 2025, Vol. 45 ›› Issue (5): 132-142.doi: 10.16708/j.cnki.1000-758X.2025.0081

• 论文 • 上一篇    下一篇

基于ICP算法的空间非合作目标超近距离实时位姿测量

赵韩雪，胡茄乾，邵长宝，江海天，李爽*   

  1. 南京航空航天大学 航天学院，南京 211106
  • 收稿日期:2024-06-03 修回日期:2024-09-07 录用日期:2024-09-11 发布日期:2025-09-17 出版日期:2025-10-01

Real time pose measurement of space non-cooperative targets at ultra-close range based on ICP algorithm

ZHAO Hanxue，HU Jiaqian，SHAO Changbao，JIANG Haitian，LI Shuang*   

  1. College of Astronautics, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China
  • Received:2024-06-03 Revision received:2024-09-07 Accepted:2024-09-11 Online:2025-09-17 Published:2025-10-01

PDF (PC)

2

摘要： 为了解决传统方法无法满足超近距离位姿测量任务中的实时性和高精度需求这一问题，提出了一种基于迭代最近点法（ICP）的空间非合作目标超近距离实时位姿测量方法。结合了针对关键帧的高精度位姿估计和针对非关键帧的实时位姿跟踪两部分。首先，针对关键帧，为了提高运算速度，采用统计滤波进行点云去噪；再通过基于体素网格重心邻近点的降采样方法进行点云稀疏。而后，利用粗配准+ICP的组合获得关键帧相对于目标模型的位姿，实现对关键帧的位姿估计。针对非关键帧，利用均匀下采样+ICP实时获得连续非关键帧之间的位姿变化，实现位姿的跟踪。最后采用关键帧的位姿估计与非关键帧之间的实时位姿跟踪相融合的方法来消除实时位姿跟踪带来的累积误差，获得高精度实时的位姿测量结果。仿真结果表明，算法可以实现对超近距离非合作目标的高精度实时位姿测量，计算频率能够达到24Hz，且姿态误差不大于1°，位置误差不大于2cm。这表明对传统ICP算法进行的全新算法架构设计，实现了针对关键帧的高精度位姿估计和针对非关键帧的实时位姿跟踪。在此基础上，将两者融合以高精度位姿估计抑制实时位姿跟踪的累积误差，在满足非合作目标超近距离测量任务高精度需求的同时保证了实时性。

关键词: 统计滤波, 迭代最近点法, 位姿测量, 空间非合作目标, 超近距离实时测量, 体素降采样

Abstract: To address the limitations of traditional methods in meeting the real-time and high-precision requirements for ultra-close-range pose measurement tasks, this paper proposes a ultra-close-range real-time pose measurement method of space non-cooperative target based on the Iterative Closest Point (ICP) algorithm. The proposed approach integrates high-precision pose estimation for keyframes with real-time pose tracking for non-keyframes. Initially, statistical filtering was applied to keyframes for point cloud denoising to enhance computational efficiency. The point cloud was then sparsified using a downsampling method based on the centroid of neighboring points within the voxel grid. Subsequently, a combination of coarse registration and ICP was employed to determine the pose of keyframes relative to the target model, thus achieving pose estimation for keyframes. For non-keyframes, uniform downsampling coupled with ICP was used to capture real-time pose changes between consecutive non-keyframes, facilitating pose tracking. Finally, the real-time pose tracking of non-keyframes was fused with pose estimation of keyframes to eliminate the cumulative errors and obtain high-precision, real-time pose measurements. Simulation results demonstrate that the proposed algorithm can achieve real-time pose measurement for close-range non-cooperative targets and its computational frequency can reach 24Hz, with an attitude error of no more than 1° and a position error within 2centimeters. These findings suggest that the proposed algorithm, designed with novel architectures based on the traditional ICP method, achieves high-precision pose estimation for keyframes and real-time position tracking for non-keyframes. The accumulative error of real-time pose tracking is suppressed by fusing high-precision pose estimation. This approach ensures real-time performance while meeting the high-precision requirements of ultra-close-range measurement tasks involving non-cooperative targets.

Key words: statistical filtering, ICP, pose measurement, noncooperative targets in space; , ultra-close-range real-time measurement, voxel downsampling