Automated Building Extraction: Comparison of Paradigms and Examples

This paper compares the paradigms of LiDAR and aerophotogrammetry in the context of building extraction and briefly discusses two roof building contour extraction methodologies. The assets and drawbacks of both data capturing system have been reported several times. In general, empirical and theoretical studies have confirmed that LiDAR methodologies are more suitable in deriving building heights and in extracting planar roof faces and ridges of the roof, whereas the aerophotogrammetry are more suitable in extracting building roof outlines. The first roof contour extraction methodology is based on a Digital Elevation Model (DEM), which is generated through the regularization of an available LiDAR point cloud. First, in order to detect aboveground objects, the DEM is segmented through a recursive splitting technique, followed by a Bayesian merging technique. The aboveground object polygons are extracted by using vectorization and polygonization techniques. Finally, the building roof contours are identified among all aboveground objects extracted previously, taking into account roof features and a Markov Random Field (MRF) model. The second methodology addresses the geometric refinement of laser-derived 3D roof contours by using high-resolution aerial images and a MRF model. First, 3D roof contours are projected onto the image-space. Then, the projected contours and the straight lines extracted from the image are used to establish an MRF description. The solution of the associate energy function provides groupings of straight lines representing roof building contours, which are topologically reconstructed based on the topology of the projected contours. Examples of applications are provided for both approaches. * Corresponding author.