No abstract

This paper proposes methods based on global descriptors to estimate the pose of a known object using a monocular camera, in the context of space rendezvous between an autonomous spacecraft and a non-cooperative target. These methods estimate the pose by detection, i.e., they require no prior information about the pose of the observed object, making them suitable for initial pose acquisition and the monitoring of faults in other on-board estimators. An approach is presented to fully retrieve the object's pose using a pre-computed set of invariants and geometric moments. Three classes of global invariant features are analyzed, based on complex moments, Zernike moments and Fourier descriptors.The robustness of the different invariants is tested under various conditions and their performance is discussed and compared. The method offers a fast and robust solution for pose estimation by detection, with a low computational complexity that is compatible with space-qualified processors.

This study addresses the issue of vision-based navigation for space rendezvous with non-cooperative targets. After a brief description of the scenario and its peculiarities, the theory underlying monocular edges-based tracking for pose estimation is recalled and an innovative tracking algorithm is formally developed and implemented. This algorithm is coupled with a dynamic Kalman Filter propagating the dynamics which underlies a space rendezvous. The navigation filter increases the robustness of target position and attitude estimation, and allows the estimation of target translational velocity and rotation rate using only pose measurements. Moreover, the filter implements a computationally efficient delay management technique that allows merging the delayed and infrequent measurements typical of vision-based navigation. The performance of the algorithm is tested in different scenarios with high fidelity synthetic images.

This paper proposes Thales Alenia Space visionbased navigation solution for close proximity operations in autonomous space rendezvous with non-cooperative targets. The proposed solution covers all the phases of the navigation. First, a neural network robustly extracts the target silhouette from complex background. Then, the binary silhouette is used to retrieve the initial relative pose using a detection algorithm. We propose an innovative approach to retrieve the object's pose using a precomputed set of invariants and geometric moments. The observation is extended over a set of consecutive frames in order to allow the rejection of outlying measurements and to obtain a robust pose initialization. Once an initial estimate of the pose is acquired, a recursive tracking algorithm based on the extraction and matching of the observed silhouette contours with the 3D geometric model of the target is initialized. The detection algorithm is run in parallel to the tracker in order to correct the tracking in case of diverging measurements. The measurements are then integrated into a dynamic filter, increasing the robustness of target pose estimation, allowing the estimation of target translational velocity and rotation rate, and implementing a computationally efficient delay management technique that allows merging delayed and infrequent measurements. The overall Navigation solution has a low computational load, which makes it compatible with space-qualified microprocessors. The solution is tested and validated in different close proximity scenarios using synthetic images generated with Thales Alenia Space rendering engine SpiCam.