Automatic Extraction of Planar Clusters and Their Contours on Building Façades Recorded by Terrestrial Laser Scanner

Since 3D city models need to be realistic not only from a bird's point of view, but also from a pedestrian's point of view, the interest in the generation of 3D façade models is increasing. This paper presents two successive algorithms for automatically segmenting building façades scanned by Terrestrial Laser Scanner (TLS) into planar clusters and extracting their contours. Since majority of façade components are planes, the topic of automatic extraction of planar features has been studied. The RANSAC algorithm has been chosen among numerous methods. It is a robust estimator frequently used to compute model parameters from a dataset containing outliers, as it occurs in TLS data. Nevertheless, the RANSAC algorithm has been improved in order to extract the most significant planar clusters describing the main features composing the building façades. Subsequently, a second algorithm has been developed for extracting the contours of these features. The innovative idea presented in this paper is the efficient way to detect the points composing the contours. In order to evaluate the performances of both algorithms, they have successively been applied on samples with different characteristics, i.e. densities, types of façades and size of architectural details. Finally, a quality evaluation based on the comparison of planar clusters and contours obtained manually has been carried out. The results prove that the proposed algorithms deliver qualitative as well as quantitative satisfactory results and confirm that both algorithms are reliable for the forthcoming 3D modelling of building façades.