Vision Navigation for Spacecraft Proximity Operations

There is a growing interest to implement computer vision systems for space applications. At present, satellites typically require large development costs and involve significant risk of failure to successfully deploy features in orbit, resulting in reduced operational capabilities or a complete malfunction. With large communications satellites costing in excess of $250 million to place in orbit, this could represent extraordinary financial loss. In developing and deploying servicer spacecraft to perform in-orbit reparations, the risk of total, functional failure of satellites could be significantly reduced. This paper proposes a vision-based navigation system for stereo camera-equipped spacecraft to model nearby target bodies to determine their relative pose and position using the Method of Least Squares and Single Value Decomposition methods. The implemented vision techniques are feature-based algorithms developed to accommodate the varying lighting of the space environment while providing foreground-background segmentation from triangulation algorithms. The algorithm was tested on a simulation of an unperturbed rotation satellite situated 30m from the observing spacecraft. The system was capable of developing a sparse 3-dimensional point cloud model as well as calculate the rotational axis - achieving the objective of estimating the relative position and pose of the target spacecraft. Development of the algorithm could provide a step towards providing automated spacecraft with the necessary situational awareness to accurately approach targets for rendezvous and docking operations as well as servicing and repair operations using simple camera setups as opposed to significantly more expensive and complicated methods based on LIDAR and similar technologies.Validerat; 20150312 (global_studentproject_submitter