For both manned and autonomous space on-orbit servicing missions, the collision avoidance and motion prediction of the target objects are essential for a safe mission operation. Our approach assumes unknown and uncooperative target objects (spacecraft, space debris) and a camera only vision system. The target's 3D structure is reconstructed from a sparse point cloud extracted from a rendezvous-SLAM algorithm and processed by a RANSAC algorithm. Afterwards the 3D model can be used in a feedback manner for enhancing visual navigation processing tasks. The paper details the general concept and presents experimental results with camera image data from a spacecraft rendezvous simulator testbed.
展开▼