Geometry compression of scanned point-clouds

摘要

Today, 3D scanners are capable of producing high resolution and accurate 3D computer models that are often used for entertainment, industrial, and scientific purposes. Because such models may contain a huge amount of points, their storage is expansive, their loading is slow, and their transmission over the internet is impossibly slow. In this paper, a method for compression of 3D scanned point clouds is presented. The method is based on a prediction schema, where four prediction rules are proposed. Three special prediction rules were developed for geometry coding, while the scalar values attached to the points (i. e. colour) are coded with a constant predictor. These geometry prediction rules exploit the scanning way of the point clouds to achieve high accuracy of the predictions. Consequentially, the prediction errors are small and therefore suitable for variable length coding. In this way, the values are broken in the strings of bytes, zero bytes are removed and nonzero bytes are formed into byte-streams. These byte-streams are compressed with arithmetic coding and stored in the output file.