Efficient Closed-Loop Detection and Pose Estimation for Vision-Only Relative Localization in Space with a Cooperative Target
Vela, Patricio A.
MetadataShow full item record
An integrated processing pipeline is presented for relative pose estimation of vision-only cooperative localization between two vehicles with unknown relative motion. The motivating scenario is that of proximity operations between two spacecraft when the target spacecraft has a special target pattern and the chase spacecraft is navigating using only a monocular visual sensor. The only prior information assumed is knowledge of the target pattern, which we propose to consist of nested circular blobs. The algorithm is useful for applications requiring localization accuracy using limited computational resources. It achieves low computational cost with high accuracy and robustness via the following contributions: (1) an adaptive visual pattern detection scheme based on the estimated relative pose, which improves both the e ciency of detection and accuracy of pose estimates; (2) a parametric blob detector called Box-LoG which is computationally e cient; and (3) an algorithm which jointly solves the frame-to-frame data association and relative pose estimation. An incremental smoothing technique temporally smooths the pose estimates. The approach can deal with target re-acquisition after loss of the target pattern from the field of view. The algorithm is tested in both synthetic simulations and on an actual spacecraft simulator platform.