Fast determination of parametric house models from dense airborne laserscanner data

In addition to its primary application field in the determination of digital elevation models, airborne laserscanning data has a great potential for modeling man-made objects such as houses. This publication shows a fast procedure for detecting and modeling buildings from raw airborne laserscanner data without the need of additional GIS or image data. Segmentation of the data is based on the computation of height texture measures, morphological filtering and local histogram analysis. The actual building model generation is based on the analysis of invariant moments applied to the segmented datasets: The parameters of standard gable roof type buildings are determined as closed solutions from ratios of binary and height-weighted moments of segmented point clouds. Deviations from the chosen building model such as dorms on a roof can be detected and modeled in a subsequent model fit analysis step. Applied to a dense laserscanner dataset with an average point density of approximately five points per square meter, the technique shows good results. Further analysis shows that the point density can be reduced to one point per square meter at a tolerable loss of accuracy. Being fast and parallelization-friendly, even an implementation of the techniques in nearrealtime systems seems feasable.