An efficient 3D point clouds outlier noise removal method based on feature consistency

The outlier noise clusters removing in 3D point cloud is a challenge due to its similar statistical distribution with the clean point cloud. This paper proposes an outlier noise clusters removing method based on feature consistency, considering that the points features of clean point clusters are similar because of the same texture and structure. First, the point cloud is partitioned into equal-sized voxel grids, and a three-dimensional index is assigned to each voxel to establish explicit adjacency relationships. Then, the points in voxel grids less than the threshold are removed. The point cloud clusters are obtained by merging the neighbor grids. Finally, the global clusters feature consistency that combines the points density and roughness are calculated, the point cloud clusters are compared with the clean points cluster by the Mahalanobis distances. The clusters whose distances exceed the setting threshold are considered noise clusters and removed. The proposed method is compared with the Euclidean clustering denoising method, the non-iterative denoising method and the POINTCLEANNET denoising method on three datasets. The results show that the denoising time of the proposed method is the shortest, the denoising precision and recall is the best.