On the Quality of Object Classification and Automated Building Modelling Based on Laserscanning Data

In this paper, techniques for an automated extraction, classification and modelling of 3D objects will be presented. They use solely laserscanning derived digital elevation models as data basis. Trees, buildings and terrain objects are detected and classified. Firstly, a special region growing algorithm segments 3D objects on the terrain surface. Object specific features are extracted inside these segments and used for classification by means of a fuzzy logic approach. Subsequently, a quality analysis is based on classification rates which are compiled in a confusion matrix. For geometric modelling of buildings a generic approach is used where buildings are approximated by (oblique) planes. The intersection of neighbouring planes leads to building edges and corners. The quality analysis of these building models is separated into positioning and height accuracy. First experimental results are presented.