An adaptive multi-scale point cloud filtering method for feature information retention

Point cloud data is often accompanied by a large number of noise and outliers, to improve the quality of point cloud data, this paper proposes a multi-scale point cloud filtering method for feature information preservation. Firstly, a multi-scale division method with distance-weighted Principal Component Analysis (DWPCA) is proposed to divide the point cloud data into large-scale outlier regions and small-scale noise regions. Secondly, we proposed a small-scale filtering method by the cosine value between the Angle of the normal vector and the tangent plane of the neighboring point used as the filtering factor of bilateral filtering for the purpose of overcoming the difficulty of retaining the details information of the multi-scale filtering. Finally, an irregular region smoothing method with cubic B-spline curve fitting and control point assignment is proposed to address the issues of key point loss and uneven edge sampling in point cloud models by redistributing control points. The experiment with shoe sole was carried out, and the results show that compared with the SOTA methods, the best average metrics appear in the proposed multi-scale filtering method, indicating that the key feature information of the point cloud model are retained and the smoothing of the scattered points in the feature area are developed by effectively removing the complex noise, thus the high-quality point cloud data are provided for 3D reconstruction.