Abstract: Aiming at the possible inaccuracy of the prior pose during the visual navigation of an asteroid landing, a feature tracking aided pose estimation method is proposed. First, the generation of navigation features relies on pre-existing pose information and a database of navigation features. Subsequently, a multi-feature discriminative correlation filter (DCF) is employed to track the position of the navigation features in the navigation camera images by combining handcrafted and depth features. The average peak correlation energy (APCE) is subsequently employed to effectively screen dependable tracking outcomes for the initial estimation of the pose. Finally, the navigation features are recalculated using the initial estimation of the pose and adjusted to match with the navigation camera image by using normalized correlation coefficients (NCC). The proposed methodology involves the integration of the process within a differentiable Levenberg-Marquardt (LM) framework, specifically designed for pose optimization. This framework incorporates constraints based on the NCC. Experimental results, utilizing images, terrain, and ephemeris data obtained from the Osiris mission, demonstrate that the proposed method's pose estimation exhibits reprojection errors within the sub-pixel range. At 1 km from the asteroid surface, the position estimation error is within 2 m and the attitude estimation error is within 1°.