Vision-based relative state estimation for a non-cooperative target

This paper presents an improved method to estimate the relative pose and motion of a non-cooperative target using stereoscopic vision. Two cameras are mounted on the chaser to acquire images of the target, and then the positions and velocities of these points in the camera frame are yielded by optical flow and photogrammetry. With the coordinates of the feature points in the target frame unknown in prior, the full dynamical states of the target relative to the chaser are estimated coarsely based on the measurements in two consecutive image frames. Then a multiplicative extend Kalman filter is designed to improve the estimation accuracy based on the former coarse results. Numerical simulations are conducted to evaluate the convergence and precision of the proposed algorithm.