To achieve six degree-of-freedom autonomous navigation of an inboard spacecraft, a novel algorithm called iterative closest imaging point (ICIP) is proposed, which deals with the pose estimation problem of a vision navigation system (VNS). This paper introduces the basics of the ICIP algorithm, including mathematical model, algorithm architecture, and convergence theory. On this basis, a navigation method is proposed. This method realizes its initialization using a Gaussian mixture model-based Kalman filter, which simultaneously solves the 3D-to-2D point correspondences and the camera pose. The initial value sensitivity, computational efficiency, robustness, and accuracy of the proposed navigation method are discussed based on simulation results. A navigation experiment verifies that the proposed method works effectively. The three-axis Euler angle accuracy is within 0.19° (1σ), and the three-axis position accuracy is within 1.87?mm (1σ). The ICIP algorithm estimates the full-state pose by merely finding the closest point couples respectively form the images obtained by the VNS and predicted at an initial value. Then the optimized solution of the imaging model is iteratively calculated and the full-state pose is obtained. Benefiting from the absence of a requirement for feature matching, the proposed navigation method offers advantages of low computational complexity, favorable stability, and applicability in an extremely simple environment in comparison with conventional methods.
展开▼
