基于点云深度学习的对称结构空间目标相对位姿测量

Relative Pose Measurement of Space Target with Symmetrical Structure Based on Point Cloud Deep Learning

针对对称结构空间目标相对位姿解算过程中点云误匹配带来的误差问题,提出一种基于点云深度学习的对称结构空间目标相对位姿测量方法。首先设计空间目标点云特征提取网络及关键点回归网络,将位姿测量问题转换为空间目标点云关键点回归问题,通过两个并行的回归网络分别输出空间目标平移向量和具有固定标签的目标点云三维边界框角点;其次利用具有连续稳定标签的角点求解目标姿态,可有效解决目标的对称结构导致的点云误配准问题;最后通过仿真数据集的实验表明,该方法相比于传统的点云配准方法有更高的准确率,能够精确求解具有对称结构的空间目标相对位姿。

Aiming at the error caused by mismatching of point cloud in the process of solving the relative pose of a space target with symmetrical structure,a method of relative pose measurement based on point cloud deep learning is proposed.Firstly,the feature extraction network and key point regression network of space target point cloud are designed,and the pose measurement problem is transformed into a key point regression problem of space target point cloud.Two parallel regression networks are used to output the space target translation vector and the 3D bounding box corner points of the target point cloud with fixed labels.Secondly,it is effective to use the corner points with continuous stable labels to solve the target attitude,which can effectively solve the point cloud mismatch problem caused by the symmetric structure of the target.Finally,the experimental results of the simulation data set show that the proposed method has higher accuracy than the traditional point cloud registration method,and can accurately solve the relative pose of the space target with symmetrical structure.