Detection and Recognition of Changes in Building Geometry Derived from Multitemporal Laserscanning Data

The work presented is part of a project that aims at change detection in urban areas after strong earthquakes. Detection and classification of these changes are used to recognise building damages as an important information input for a disaster management system based on GIS techniques. Airborne laser scanning data was chosen for this approach, because of specific advantages like data acquisition of large areas in relatively short time or extensive independence on weather and lighting conditions. Modifications are detected by comparison of digital surface models (DSMs) acquired at two different dates (t1 and t2). An analysis of solely a differential DSM (DSM(t1) DSM(t2)) would lead to ambiguities, e.g. attachments or modifications of buildings could not be related to the affected buildings. Therefore, firstly a segmentation procedure based on a region growing algorithm is used to generate separate 3D objects. For each segment object-oriented features like border gradients or shape are extracted to classify into 'building', 'vegetation' or 'terrain'. After elimination of all non-building objects, the correspondence between the objects of the two laser scanning data sets has to be determined. At this step, new and teared-off buildings are extracted. The remaining 3D objects have to be controlled in terms of significant elevation changes and thus classified into ‘not-altered’, ‘added-on’ or ‘reduced’. First results in test area 'Karlsruhe' (approx. 8km x 2km) will be presented.