Abstract:

Aiming at pose (relative attitude and position) estimation of non-cooperative spacecraft, an algorithm is developed based on monocular vision and some natural feature points.Considering the increasing estimated error caused by using the natural features,this paper introduces an iterative solution based on convex relaxation optimization and LMI algorithm to solve this problem.The optimization model in this paper is built on adverse projection. First, using relaxation algorithm, turn the non-convex and equality constrained attitude matrix to a convex and inequality constrained matrix. Then, this paper can prove that the convex problem is equal to the original problem. That is, when the convex problem gets extremum, the attitude matrix still satisfies the original equality and non-convex constrain. To further simplify this problem, we can express the convex and unequal constrain as linear matrix inequalities. At last, we can solve it with the developed interior point method and prove convergence of this algorithm. Finally, in the background of on-orbit servicing, the simulation experiment shows that this algorithm can converge within 7 iterations. Compared with SVD, this algorithm nearly doubles accuracy when noise increases gradually. And the results show that this algorithm is robust and efficient.

Key words: On-orbit servicing, Monocular vision, Pose estimation, Adverse projection, Convex relaxation optimization