Denoising Point Cloud Based on the Connexity of Spatial Grid

Abstract

To ensure no loss of feature information of model and rapidly eliminate the high density noise clusters of point cloud. A kind of de-noising algorithm based on the connexity of spatial grid was proposed. Firstly, the concept of connected region in the binary image was extended to 3D point cloud model, divided the point cloud model into uniform space hexahedron, created the concept of adjacent lattice cell, then determined the pixel value of cell grid according to the density of the point in it, and the connected area was established by matching the pixel values of the adjacent units, judged whether the lattice cell was a noise unit according the size of connected region, so that the large scale noise was filtered. Finally, used the bilateral filtering to smoothing the small-scale noise. The experimental results show the proposed method could filter out the noise while preserving the details feature of model, the maximum error of point cloud is 1.7245E−2 mm, and standard deviation is 0.4012E−2 mm after filtering, the computation time is 53 s far less than that of the clustering algorithm; under the interference of different intensity noise, the proposed algorithm can achieve better de-noising effects and retain the high-frequency feature information of point cloud.