基于单目视觉的空间非合作目标相对姿态估计方法

Monocular vision-based relative attitude estimation for non-cooperative space targets

提出一种基于单目视觉的连续多帧空间非合作目标姿态估计算法.根据目标主体主轴方向和太阳能帆板方向将目标姿态切分成姿态子空间,采用方向Chamfer匹配在姿态子空间中搜寻与目标匹配的候选姿态.通过连续多帧候选姿态的方向Chamfer匹配代价和转移代价来构建轨迹的总代价,采用动态规划对轨迹代价进行优化.结果表明该算法求解精度高,鲁棒性好.

In order to minimize the interference of solar array on attitude estimation,a novel relative attitude estimation approach based on monocular vision was proposed for non-cooperative space targets. Space targets were firstly divided to post-subspace of retrievable and irretrievable solar array,then reduce the size of the pose-subspaces by the principal axis of the target and solar array.Subsequently,fast direction Chamfer matching was adopted to obtain the matched pose with the same shape of frame from the subspace pose dataset,which was followed by being recovered to a continuous pose sequence by dynamic programming. Experimental results showed that standard deviations of rotation angle,and based on the proposed approach were all lower than those of Chamfer method and feature-based method. To summarize,the proposed approach has high accuracy and robustness.