A comprehensive framework of building model reconstruction from airborne LiDAR data

Abstract

This paper presents a comprehensive framework of reconstructing 3D building models from airborne LiDAR data, which involves building extraction, roof segmentation and model generation. Firstly, building points are extracted from LiDAR point clouds by removing walls, trees, ground and noises. Walls and trees are identified by the normal and multi-return features respectively and then ground and noise are detected by the region growing algorithm which aims at extracting smooth surfaces. Then the connected component analysis is performed to extract building points. Secondly, once the building points are acquired, building roofs are separated by the region growing algorithm which employs the normal vector and curvature of points to detect planar clusters. Finally, by combining regular building outlines obtained from building points and roof intersections acquired from the roof segmentation results, 3D building models with high accuracy are derived. Experimental results demonstrate that the proposed method is able to correctly obtain building points and reconstruct 3D building models with high accuracy.