LAUSR

This article proposes a vision-based framework to track the pose of lander in real time during planetary exploration with craters as landmarks. The contour of landmark crater is represented with three-dimensional Fourier series offline. During tracking, for the first instant, the tracking system is initialized by crater-based correspondence and optimization. For each subsequent instant, with the initial guess from the extended Kalman filter (EKF), the lander pose is determined by L1-norm minimization of the reprojection errors of the crater contour models in the descent image. The covariance of the determined pose is inferred based on Laplace distribution. With this covariance, the EKF generates the final estimate to pose and gives the initial guess for the pose at the next instant. Sufficient trails verify the efficacy of the proposed method.