The classification of airborne LiDAR building point clouds based on multi‐scale and multi‐level cloth simulation

Abstract

AbstractBuilding point clouds classification is an important task for a Smart City. This paper proposes a method based on multi‐scale and multi‐level cloth simulation to classify buildings in airborne LiDAR point clouds. First, the point clouds are preprocessed to denoise, filter ground and normalise height. The height distribution of point clouds in the whole area is then counted. Point clouds are segmented with similar heights on the top of the internal buildings by using the watershed algorithm based on height hierarchical labelling. Finally, the multi‐scale cloth simulation is used to crudely extract the seed points on the top of the buildings. These seed points are grown to realise the complete classification of airborne building point clouds. Multiple sets of data are selected for experimental verification. The results show that this method has higher classification accuracy and better applicability with a correct classification rate of 96.83%, 95.94% and 94.36%, and a recall rate of 98.18%, 96.32% and 92.33% in point level.