Hierarchical Bayesian nets for building extraction using dense digital surface models

During the last years an increasing demand for 3D data of urban scenes can be recognized. Techniques for automatic acquisition of buildings are needed to satisfy this demand in an economic way. This paper describes an approach for building extraction using digital surface models (DSM) as input data. The first task is the detection of areas within the DSM which describe buildings. The second task is the reconstruction of geometric building descriptions. In this paper we focus on new extensions of our approach. The first extension is the detection of buildings using two alternative classification schemes: a binary or a statistical classification based on Bayesian nets, both using local geometric properties. The second extension is the extraction of roof structures as a first step towards the reconstruction of polyhedral building descriptions. (C) 1998 Elsevier Science B.V. All rights reserved.