Data Reduction of Indoor Point Clouds

摘要

The reconstruction and visualization of three-dimensional point-cloud models, obtained by terrestrial laser scanners, is interesting to many research areas. This paper presents an algorithm to decimate redundant information in real-world indoor point-cloud scenes. The key idea is to recognize planar segments from the point-cloud and to decimate their inlier points by the triangulation of the boundary, describing the shape. To achieve this RANSAC, normal vector filtering, statistical clustering, alpha shape boundary recognition and the constrained Delaunay triangulation are used. The algorithm is tested on various large dense point-clouds and is capable of reduction rates from approximately 75-95%.