Addressing the issues of point cloud simplification algorithms that rely on traditional parameters when extracting features, which is not comprehensive and easy to lose feature boundaries, this study provides a point cloud simplification approach using von Mises-Fisher (vMF) distribution to extract features. This method first uses a neighborhood center point to create a vector, divides the surface through the threshold of the relationship with the normal direction, reduces the impact of noise on the finer features. Then, the priority of surface points is extracted by using vMF distribution to realize global feature extraction. Finally, octree hierarchical simplification is operated based on features. Experiments described that the method in this study can successfully extract detailed features. Compared with methods based on curvature and Hausdorff distance, it has a better feature extraction effect. The simplification algorithms based on curvature, grid, and random, and the proposed method are used to analyze the reconstruction results, 3D bias, and quantitative analysis, results prove that the proposed simplification method is more effective. The proposed simplification method provides a fresh approach for point cloud feature extraction and simplification.