Aiming at the problem that the conventional point cloud filtering method will cause greater damage to the model in the process of removing the noise close to the model, a point cloud filtering algorithm combining dual tensor voting and multi-scale normal vector estimation is proposed. First, the principal component analysis method is used to estimate the normal vector of each point on a larger scale, and the double tensor voting is performed on each point to extract the feature points. Then, the normal vectors of the extracted feature points are estimated at a smaller scale, and the small-scale noise plane is eliminated by combining the random sample consensus method. Finally, the curvature is used to filter the remaining noise to obtain the final point cloud data. Experimental results show that the proposed algorithm can effectively eliminate noise points, and better retain the sharp features of the 3D model, which lays the foundation for subsequent point cloud registration and 3D reconstruction.