A low-dimensional binary-based descriptor for unknown satellite relative pose estimation

Abstract

Satellite pose estimation is an essential task for on-orbit service. After obtaining sequential images of the unknown satellite, its relative pose can be estimated by capturing and matching keypoints in these images. Generally, existing point-based methods will produce numerous keypoints, represented by high-dimensional descriptors. Consequently, they are inappropriate to estimate the relative pose between the observer and the unknown target, due to the limited capacity of storage and computation of the on-board processor. On this basis, a low-dimensional binary-based descriptor is proposed, in conjunction with the use of a keypoint selection algorithm, to address the above-mentioned limitation. The developed algorithm is then used to select the keypoints from a series of points generated by the Harris detector. Each keypoint is represented by a novel 72-dimensional binary descriptor, the vicinity of which is divided into 12 sectors, and then each sector is mathematically encoded by six binary bits. The experimental results obtained via the presented approach demonstrate a good coincidence with that of the SURF and ORB methods, in terms of the keypoint matching and pose estimation accuracies, while it consumes less storage and has a faster matching speed.