According to the principle of 3D reconstruction of binocular stereo vision, if the dense 3D point cloud data can be obtained, the characteristics of 3D object surface shape can be determined. Therefore, this paper proposes an improved image registration method of BAS. Firstly, SURF algorithm is used to extract the feature points of the image taken by binocular camera, and the feature vector and its corresponding position are extracted from the returned binary or intensity image. Finally, the relationship matrix between the two images is taken as the objective function of the improved BAS to calculate the transformation relationship between left and right images in binocular stereo vision. The two-dimensional point cloud data of the inner surface in one of the images is found by image segmentation, and the corresponding two-dimensional point cloud data is calculated by using the relation matrix, and finally the three-dimensional point cloud data of the inner side is reconstructed. From the comparative analysis of the images, we can find the corresponding points to calculate the relationship matrix, so that the SURF algorithm of image registration can be more accurate by using the rotation translation matrix calculated in this paper.