Adaptive random sample consensus approach for segmentation of building roof in airborne laser scanning point cloud

摘要

This work proposes a three-step method for segmenting the roof planes of buildings in Airborne Laser Scanning (ALS) data. The first step aims at mainly avoiding the exhaustive search for planar roof faces throughout the ALS point cloud. Standard algorithms for processing ALS point cloud are used to isolate building regions. The second step of the proposed method consists in segmenting roof planes within building regions previously delimited. We use the RANdom SAmple Consensus (RANSAC) algorithm to detect roof plane points, taking into account two adaptive parameters for checking the consistency of ALS building points with the candidate planes: the distance between ALS building points and candidate planes; and the angle between the gradient vectors at ALS building points and the candidate planes' normal vector. Each ALS building point is classified as consistent if computed parameters are below corresponding thresholds, which are automatically determined by thresholding histograms constructed for both parameters. As the RANSAC algorithm can generate fragmented results, in the third step, a post-processing is accomplished to merge planes that are approximately collinear and spatially close. The results show that the proposed method works properly. However, failures occur mainly in regions affected by local anomalies such as trees and antennas. Average rates around 90% and higher than 95% have been obtained for the completeness and correction quality parameters, respectively.