Abstract: Selection and reduction for large point clouds are the indispensable
post-processing steps to deal with the problem of background data, redundant sampling and
non-uniform distributions. Traditional methods still have many limitations for custom low cost 3D
scanning system. There is no public available point cloud selection algorithm supporting lasso UI
interfaces in research community; traditional point cloud reduction algorithm focuses on adapting
samples to surface curvature without considering balanced distribution in flat regions. This paper
presents a new point cloud selection algorithm with lasso UI interfaces, which avoids most of the
point-in-polygon tests by constructing rectangle covering of the lasso polygon and the octree
encoding of the input point cloud. Based on Poisson-disk sampling, a novel point cloud reduction
approach is proposed, in which the radius of the disk embedded on surface is defined by Boolean
intersection between neighborhood spheres of samples resulting in the property of sharp edge
feature preserving. The experimental results demonstrate that these methods provide satisfactory
results for point cloud removal preprocessing in low cost 3D scanning system.

Key words: point cloud selection, point cloud reduction, Poisson-disk sampling, sharp edge
feature preserving