Clustering irregular spaced lidar TINs for 3D reconstruction

Abstract

Several sets of features, existent in triangulated, irregularly spaced LiDAR data, are extracted, conditioned, and presented to a number of clustering algorithms with the intent to recognize planar structures within the data. From those planar structures, encoded by the clustering algorithms, 3D models are then reconstructed. The purpose of this paper is to evaluate the performance of these clustering algorithms' ability to accurately cluster coplanar triangles into groups correlating to a given, depicted structure's roof planes. Several preprocessing, input conditioning procedures are presented. Also, a post processing planar regression algorithm is implemented to further refine the clustering algorithms' results to realize 3D reconstructed models of the LiDAR points. Furthermore, membership criterions, for a given triangle to correctly belong to a roof cluster, are proposed. Measures in which to evaluate the performance of the clustering algorithms ability to accurately encode the triangulated LiDAR data are also proposed. Copyright © 2008 by the International Institute of Informatics and Systemics.