High noise and redundancy of outliers in the initial point cloud result in low efficiency of three-dimensional reconstruction and rough surface of reconstructed surface. Thus, this study proposes an improved preprocessing algorithm for an adaptive simplified point cloud. Statistical filtering was used to eliminate outlier noise, and the hyperbolic tangent function was introduced into the downsampling based on the voxel center of gravity adjacent feature points to maintain the point cloud features and simplify the point cloud data. The moving least square fitting function was then established. Furthermore, its quadratic basis function and Gaussian weight function were determined, and the point cloud data was smoothed and optimized. Finally, the projection triangulation algorithm was used to reconstruct the point cloud surface. Experimental results show that the proposed algorithm can effectively remove outliers, simplify point cloud data, and improve the efficiency of surface reconstruction, and the reconstructed model has a smooth surface and fewer holes.