中国空间科学技术

中国空间科学技术 ›› 2022, Vol. 42 ›› Issue (4): 36-44.doi: 10.16708/j.cnki.1000-758X.2022.0050

• 论文 • 上一篇    下一篇

基于单目视觉的非合作目标相对位姿测量

邢艳军*，王浩，叶东，张加迅   

  1. 1北京空间飞行器总体设计部，北京100094   2哈尔滨工业大学 卫星技术研究所，哈尔滨150000
  • 出版日期:2022-08-25 发布日期:2022-08-09

Relative pose measurement for non-cooperative target based on monocular vision

XING Yanjun*，WANG Hao，YE Dong，ZHANG Jiaxun   

  1. 1 Beijing Institute of Spacecraft System Engineering,Beijing 100094,China   2 Rearch Center of Satellite Technology，Harbin Institute of Technology,Harbin 150000,China
  • Published:2022-08-25 Online:2022-08-09

PDF (PC)

340

摘要： 为实现对空间姿态翻滚航天器的在轨服务与维护以及对空间碎片的清理，需对其进行精确的相对位姿测量。针对相对位姿测量问题，提出了基于单目视觉与卡尔曼滤波的相对位姿测量方法。通过对特征点匹配算法进行调查，采用了具有尺度不变性与旋转不变性的尺度不变特征变换算法(SIFT)和加速稳健特征算法(SURF)的特征点提取方法，并对二者进行了对比，得到了二者分别适用的工况条件。通过对Kalman滤波算法进行研究，引入了相机偏置矩阵，设计了Kalman滤波器，解决了单目相机的距离模糊问题，估计得到了非合作目标的相对位姿、主惯量比以及特征点位置信息。经过仿真，姿态角度估计误差在稳定后低于0.3°，相对位置估计误差在稳定后低于0.5m，相较于真值，误差小于1.67%，主惯量比估计误差在稳定后低于0.01，特征点位置误差在稳定后低于0.005m。在引入相机偏置条件后，滤波状态变量均收敛，并得到具有足够精度的估计，成功解决了单目相机深度信息缺失问题。

关键词: 单目视觉, 非合作目标, SIFT, SURF, Kalman滤波, 相对位姿估计

Abstract: In order to realize the on orbit service and maintenance of the space attitude tumbling spacecraft and the removal of space debris,it is necessary to perform accurate relative pose measurement.Aiming at the problems,a relative pose measurement method based on monocular vision and Kalman filter was proposed.By investigating the feature point matching algorithm,the feature point extraction methods based on the scale invariant feature transform (SIFT) algorithm and the speeded up robust feature (SURF) algorithm with scale invariance and rotation invariance were used.And these two algorithms were further compared to get the working conditions of each other.Through the study of Kalman filter algorithm,the camera bias matrix was introduced,the Kalman filter was designed,the range ambiguity problem of the monocular camera was solved,and the relative pose,main inertia ratio and feature point position information of non-cooperative targets were estimated.According to the simulation,the attitude angle estimation error is less than 0.3° after stabilization,and the relative position estimation error is less than 0.5m.Compared with the true values,the errors are less than 1.67%.The main inertia ratio estimation error is less than 0.01,and the feature point position error is less than 0.005m after stabilization.After introducing the camera bias condition,all the filtering state variables converged,and an estimation with sufficient accuracy was obtained.The problem of the lack of depth information of the monocular camera has been successfully solved.

Key words: monocular vision, non-cooperative goal, SIFT, SURF, the Kalman filter, relative pose estimation