Performance Evaluation of 3D Model-based Techniques for Autonomous Pose Initialization and Tracking

This paper deals with pose estimation of an uncooperative target by exploiting 3D point clouds which can be provided by active LIDARs or passive stereovision systems. This problem is relevant to Formation Flying, On-Orbit Servicing, and Active Debris Removal missions characterized by close-proximity flight phases. Target relative position and attitude are initialized by an original on-line three-dimensional template matching technique, while different variants of the Iterative Closest Point algorithm are considered for the tracking phase. A numerical simulation environment has been developed for algorithm performance evaluation. It includes realistic debris geometry, target-chaser relative dynamics, and sensor operation. Results relevant to a large debris in Low Earth Orbit show that the proposed template matching approach is robust against highly variable conditions in terms of target pose while simultaneously not demanding in terms of computational cost and on-board data storage. As regards the tracking stage, the simulations prove that the analyzed techniques are able to provide sub-degree accuracy in attitude and sub-centimeter accuracy in the relative position, even when dealing with very sparse point clouds.