Selecting Key Features of 3D Object Model for Relative Pose Estimation

Abstract

This paper addresses the discussion of the complexity and level of detail of the 3D model of a target satellite that is required in order to get a stable and accurate relative pose estimation with monocular cameras during an On-Orbit Servicing (OOS) mission scenario. We assume to know the target 3D mesh with thousands of vertices before the mission takes place. Nevertheless, for on-board pose estimation we have to use compact 3D models of the targets with only some key points. In this paper we compare the pose estimation results using three models for tests. The first model has manually extracted key features, where the second one includes only features extracted with Harris3D technique. The third model includes a part of Harris3D key features and some manually selected points. The offline pose estimation tests were using the data from European Proximity Operations Simulator (EPOS) facility.