期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2010
卷号：XXXVIII - Part 3B
页码：126-130
出版社：Copernicus Publications
摘要：This paper focuses on extracting vertical objects from 3D terrestrial point clouds, acquired in dense urban scenes. We are especially interested in urban vertical posts whose inventory is useful for many applications such as urban mapping, virtual tourism or localization. The proposed methodology is based on two steps. The first is a focalization step providing hypothetical candidates consisting of vertical features. The second step validates or rejects these candidates. In the case of validation, the features are classified according to the posts pattern library. The extraction of vertical objects is processed by projecting the point cloud in a horizontal plan. The accumulation density, the minimal and maximal heights are used to filter out ground points, tree leaves and to solve acquisition problems such as a region multiscans. After filtering step, the accumulation image is updated. Connex regions correspond to the vertical objects. These candidates are then validated regarding the posts pattern library. An analysis of the 3D vertical point distribution is processed. To do that, each region is characterized by its eigenvalues and eigenvectors based on a Principal Component Analysis in 3D space. The classification is processed by a decision tree algorithm. Results are presented on large and various datasets acquired under real conditions in a dense urban area. A global accuracy of 84 % is reached
关键词：Terrestrial Laser scanning; urban feature extraction; vertical post; decision tree
