Building segmentation and modeling from airborne LiDAR data

Abstract

Due to the high accuracy and fast acquisition speed offered by airborne Light Detection and Ranging (LiDAR) technology, airborne LiDAR point clouds have been widely used in three-dimensional building model reconstruction. This paper presents a novel approach to segment building roofs from point clouds using a Gaussian mixture model in which buildings are represented by a mixture of Gaussians (MoG). The Expectation-Maximization (EM) algorithm with the minimum description length (MDL) principle is employed to obtain the optimal parameters of the MoG model for separating building roofs. To separate complete planar building roofs, coplanar Gaussian components are merged according to their distances to the corresponding planes. In addition, shape analysis is utilized to remove nonplanar objects caused by trees and irregular artifacts. Building models are obtained by combining segmented planar roofs, topological relationships, and regularized building boundaries. Roof intersection segments and points are derived by the segmentation results, and a raster-based regularization method is employed to obtain geometrically correct and regular building models. Experimental results suggest that the segmentation method is able to separate building roofs with high accuracy while maintaining correct topological relationships among roofs.