In order to improve the matching efficiency and reconstruction quality of binocular feature matching in 3D reconstruction, this paper proposed a SURF-RANSAC based on the traditional speed up robust features (SURF) matching algorithm,which was use the left and right images for 3D reconstruction. Firstly, the initial feature points of the target image are obtained by Hessian matrix, and the preliminary feature point matching is completed by the fast library for approximate nearest neighbors algorithm, and then merged with the random?sample?consensus (RANSAC) to optimize the matching, and finally use 3D coordinates and texture mapping to complete the 3D reconstruction. In this paper, the algorithm is verified on Open CV. The results show that the proposed algorithm has higher accuracy and faster speed than the traditional 3D reconstruction algorithm.