Abstract: A method for 3D shape reconstruction of small celestial body based on sequence images is proposed.Using the internal geometric constraints of sequence images,the relative position and attitude parameters ofsequence images are estimated based on sparse bundle adjustment.In order to reconstruct the 3D shape of the small celestial body,such solutions as epipolar line geometric constraints,semi-global matching strategy and multi-view least squares matching are combined to perform dense image matching.In order to solve the mis-matching problem caused by the different scale and lack of texture details,the search strategy of random KD tree and RANSAC method with perspective model are used to detect outliers.Then,the 3D shape of VESTA is reconstructed using the method above by sequence images obtained from the Dawn mission.The results demonstrate our method’s effectiveness.