A Novel Approach to Visual Navigation based on Feature Line Correspondences for Precision Landing

To satisfy the needs of precise pin-point landing missions in deep space exploration, this paper proposes a method based on feature line extraction and matching to estimate the attitude and position of a lander during the descent phase. Linear equations for a lander's motion parameters are given by using at least three feature lines on the planetary surface and their two-dimensional projections. Then, by taking advantage of Singular Value Decomposition (SVD), candidate solutions are obtained. Lastly, the unique lander's attitude and position relative to the landing site are selected from the candidate solutions. Simulation results show that the proposed algorithm is able to estimate a lander's attitude and position robustly and quickly. Without an extended Kalman filter, the average errors of attitude are less than 1(degrees) and the average errors of position are less than 10 m at an altitude of 2,000 m. With an extended Kalman filter, attitude errors are within 0.5(degrees) and position errors are within 1 m at an altitude of 247.9 m.