改进PointNetLK的点云智能配准与位姿图优化方法

针对空间在轨服务任务中的非合作目标相对位姿测量问题，提出一种目标可测部位点云的智能配准方法。首先，通过Straight Through滤波算法对半物理仿真平台采集得到的点云进行目标提取，以消除背景数据等杂乱信息；其次，改进PointNetLK神经网络点云配准算法，将提取后的点云数据作为输入，从而获得初步配准结果，解决非合作目标先验信息缺失导致的无法配准问题；最后，建立基于位姿图的优化模型，以降低配准误差，提高配准精度。实验结果表明，与传统迭代最近点(ICP)算法相比，配准综合误差从6.3598降低到1.7291，精度提高约 72.81% 单次耗时从33.16 s降低到4.2 s，效率提升约87.33%，与当前SM ICP等其他算法相比，也具有一定的优势。

Improved PointNetLK Method for Point Cloud Intelligent Registration and Pose Graph Optimization

Aiming at the problem of relative pose measurement of non cooperative target in space on orbit servicing tasks, an intelligent registration method of target measurable point cloud is proposed. Firstly, the point cloud collected by semi physical simulation platform is extracted by the Straight Through filtering algorithm to eliminate clutter information such as background data. Secondly, the point cloud registration algorithm of PointNetLK neural network is improved, and the extracted point cloud data is used as input to obtain preliminary registration results, and the problem of non cooperative target registration failure caused by the absence of prior information is solved. Finally, an optimization model based on the pose graph is established to reduce the registration error and improve the registration accuracy. Experimental results show that compared with the traditional iterative closest point (ICP) algorithm, the comprehensive error of registration is reduced from 6.3598 to 1.7291, and the accuracy is improved by about 72.81%. The single time consumption is reduced from 33.16 s to 4.2 s, and the efficiency is improved by about 87.33%. Compared with the current SM ICP algorithm and other algorithms, it also has certain advantages.