DETECTION OF COLLAPSED BUILDINGS BY CLASSIFYING SEGMENTED AIRBORNE LASER SCANNER DATA
Keywords: Supervised classification, maximum entropy modelling, rule based classification, airborne laser scanner data, segmentation, object-based point cloud analysis
Abstract. Rapid mapping of damaged regions and individual buildings is essential for efficient crisis management. Airborne laser scanner (ALS) data is potentially able to deliver accurate information on the 3D structures in a damaged region. In this paper we describe two different strategies how to process ALS point clouds in order to detect collapsed buildings automatically. Our aim is to detect collapsed buildings using post event data only. The first step in the workflow is the segmentation of the point cloud detecting planar regions. Next, various attributes are calculated for each segment. The detection of damaged buildings is based on the values of these attributes. Two different classification strategies have been applied in order to test whether the chosen strategy is capable of detect- ing collapsed buildings. The results of the classification are analysed and assessed for accuracy against a reference map in order to validate the quality of the rules derived. Classification results have been achieved with accuracy measures from 60–85% complete- ness and correctness. It is shown that not only the classification strategy influences the accuracy measures; also the validation meth- odology, including the type and accuracy of the reference data, plays a major role.