Novel pose measurement with optimized principal component analysis for unknown spacecraft based on point cloud

This paper investigates the issue of vision orientation for unknown spacecraft in orbit-capture, upon which a fast and highly accurate pose measurement method based on improved coordinate system correction by weighted principal component analysis (PCA) is proposed. This algorithm weights point cloud features before dimensionality reduction, and then three principal component vectors in different frames are calculated. Consequently, the effective reduction of the original point cloud and the reduction of information overlap are achieved. The nearest point of the Euclidean distance is employed to corrected the direction of PCA coordinate axis, and thus the initial pose of two sets of point cloud are obtained. Finally, the point cloud in arbitrary pose relationship of unknown space can be aligned accurately by improved iterative closest point (ICP) algorithm with the kd-tree search strategy. The presented method overcomes the disadvantages of high requirement of initial value and avoiding local convergence, which means it achieves a global alignment for unknown target with point cloud of similar shape and integrity. Experiments show that the maximum relative error of attitude is superior to 0.15°, position error is less than ±4mm within the space 2000mm×2000mm×3000mm. Results verify that the accuracy and speed performance of the proposed approach can satisfy the requirements of on-orbit spacecraft to capture unknown objects.