Research on Key Technologies of Spatial Non-Cooperative Target Pose Measurement

For the pose estimation problem of spatial non-cooperative targets, a point cloud normal vector and curvature key feature registration algorithm is proposed. The curvatures under different neighborhood radii are calculated through principal component analysis, key points are selected, and the initial point cloud data is downsampled. For each key point, a seven-dimensional feature descriptor composed of 4 normal vectors and three curvatures is adopted to study the angle and curvature features of multiple neighborhood radii. The similarity of key point feature descriptors between the source point cloud and the target point cloud was analyzed. The ratio of the Euclidean minimum distance to the sub-minimum distance determines the initial correspondence between points. The Random Sample Consensus (RANSAC) algorithm is adopted to eliminate incorrect correspondences. To achieve better registration performance, the Iterative Closest Point (ICP) algorithm was introduced for fine registration. The transformation matrix was calculated, the registration error was analyzed, and a Gaussian noise suppression analysis was performed. The experimental results show that the root mean square error of the feature extraction and registration experiment for space satellite targets is 2.71mm, and the attitude angle error in the y direction is 0.427°. This method has extensive application value in the pose measurement of space targets.