Point cloud denoising algorithm with geometric feature preserving

Fuqun Zhao

doi:10.1007/s00530-022-00936-4

Return

Point cloud denoising algorithm with geometric feature preserving

2022-04-29

PRE

Fuqun Zhao *

DOI：10.1007/s00530-022-00936-4

Original

Original request for help

Save

Abstract

En 中文

Point cloud data model can be obtained by 3D laser scanning technology, but due to the roughness, surface texture and measurement environment of the object, the original point cloud data model usually contains a large number of noise points. Aiming at the problem that the complex noises in point cloud data model are difficult to remove, a point cloud denoising algorithm with geometric feature preservation is proposed. Firstly, the tensor voting matrix of points and their neighboring points in point cloud was calculated, and the eigenvalues and eigenvectors of the matrix were solved; secondly, the diffusion tensor based on the eigenvalues and eigenvectors was constructed, and the anisotropic diffusion equation based on the diffusion tensor was adopted to filter the noises so as to realize the initial denoising of point cloud; thirdly, the curvature factor and the density of data points in the initial denoising point cloud were defined, and the objective function of fuzzy c-means (FCM) clustering method was constructed by weighting the degree factor. Finally, the feature weighted FCM algorithm was used to further delete the noises in the point cloud, and the final accurate denoising was achieved. The experimental results show that the geometric feature preserving denoising algorithm has good denoising effect on both public point cloud, cultural relic point cloud and outdoor point cloud, so the proposed denoising algorithm is an effective point cloud denoising algorithm.

Keywords:

Point cloud denoising

Tensor voting

Fuzzy c-means clustering

Curvature

Density