Weighted Sample Elimination: Subsampling Point Clouds Using Poisson Disk Sampling

This research focuses on developing efficient algorithms for subsampling point clouds using Poisson disk sampling techniques. Point clouds are unorganized sets of points in 2D or 3D space that represent surfaces of objects, scenes, or areas. These point clouds can consist of hundreds of thousands to millions of points, so reducing their size can make numerically analyzing their properties much more efficient. This reduction, called subsampling, aims to create a smaller set of points that preserves the essential features of the original cloud. An ideal subsampling method would maximize the minimum distance between points, achieving what is known as the maximal Poisson disk radius. However, this is computationally intractable (NP-hard). Approximate solutions, known as Poisson disk sampling algorithms, can be used to combat this issue. This research specifically focuses on one of these methods, called Weighted Sample Elimination. My project implemented this algorithm and extended it for subsampling point clouds with anisotropic distributions, where point spacing varies depending on the local geometry. This extension could improve efficiency and accuracy in applications with complex spatial data.