Realizing Uniformity of 3D Point Clouds Based on Improved Poisson-Disk Sampling

Abstract

With the development of digital technology, large-scale three-dimensional point cloud data are formed from data obtained by output from three-dimensional (3D) laser measurements and numerical fluid simulations. The complex shape described by 3D laser measurements is utilized in the project of digital archiving, which is an effort to leave cultural property to future generations. The data obtained by numerical fluid simulation is also useful for analyzing the dynamic behavior of the ocean. For utilization and analysis of these point clouds, visualization is important as visual support. The quality of visualization depends on the uniformity of the point distribution. However, in the case of a 3D point cloud obtained by measurement or fluid simulation, bias of point density may occur in the point distribution due to the measurement environment or the process of converting to a point cloud. This will impair the visualization quality. In previous studies, we used Poisson disk sampling (PDS) to eliminate point distribution bias and improve visualization quality. However, in point reduction by the naive PDS, since point selection is randomly performed in processing, the uniformity of the inter-point distance between points is insufficient. In this paper, we propose “dual-shell PDS” as a method to improve PDS and generate point clouds with constant distance between points and improved visualization quality.