Building area detection based on the perceptual cues of surface patches extracted from airborne LIDAR point clouds

In this study, we propose a fast approach to detect building areas using airborne LIDAR data. This approach consists of four main steps. First, we extract non-ground points from the LIDAR point cloud using a filtering method. We then generate surface patches from the extracted points using a robust segmentation method based on adaptive region growing processes. Each patch is specified with the planar parameters with the fitting error, all the assigned points to the patch, and the boundary points among them. We also compute the properties of each patch such as the area, roughness, mean height above a digital terrain model. Finally, based on such patch properties, we determine whether each patch belongs to a building or not. The detected building areas were compared with the real building areas manually detected from the aerial images corresponding to the LIDAR data. The comparison result shows that the proposed algorithm can classify building areas successfully with the detection rate of 83%.