Structure from Motion has been widely used in 3D reconstruction based on multiview geometry, which is used to construct sparse point clouds and solve camera pose information. However, there are often problematic matching pairs in image sets, which makes it difficult to obtain correct reconstruction results by SFM method. Aiming at the problems of repetitive structure and short baseline, this paper constructs the Minimum Spanning Tree to filter the mismatched image pairs, and reconstructs the connection relationship between images, thus improving the incremental SFM method. In the initialization stage, the edge weights are calculated, and the MST is extracted, and then iterative merging is performed to get the final image connection graph from which the mismatching is filtered, and then it is expanded in an incremental way until the images in the whole data set are traversed. The evaluation results of public data sets and data sets with repeated structures show that the improved incremental SFM method can achieve similar or better reconstruction results.