A novel approach to visual navigation based on multi-feature matching

Abstract

In order to meet the requirement of autonomous navigation in deep space exploration, this paper presents a novel visual navigation method. The visual navigation algorithm based on single feature matching is a common visual method to calculate the attitude and position of a lander. However, the algorithm based on feature line matching or crater matching is great limited due to the feature line extracted and matching is easy to be influenced by noises or the craters are sparse on the surface of some planets. To solve these problems, the paper designs an algorithm to estimate the lander's motion parameters based on multi-feature, feature line and crater, matching. Firstly, the preliminary estimation of attitude and position of a lander is got via using Perspective-n-Line based on feature line matching and Kronecker product based on crater matching respectively. Then, the final motion parameters are estimated according to the parallel extended Kalman filter of information fusion technology. Simulation results show that the position errors of our algorithm are less than 1m and the attitude errors of our algorithm are less than 0.5° at altitude of 247.9m.