Extraction of building roof planes with stratified random sample consensus
MetadataShow full item record
This paper describes a consensus-set estimation for building roof-plane detection using a stratified random sample consensus (sRANSAC) algorithm applied to point clouds acquired by laser scanning systems. The main idea is to use one initial classification to generate consensus-set candidates to optimise the sampling mechanism compared to the original RANSAC. The initial classification is performed using mathematical morphology to filter ground returns and estimate local variance information to detect potential planar regions. Thus, the algorithm can prioritise points within planar segments and the number of iterations can be estimated dynamically from available data. The results based on experiments using five different lidar datasets indicate that the proposed method reduces the number of computations for building roof-plane detection and also improves accuracy compared to RANSAC.