A novel technique for building roof mapping in very-high-resolution multispectral satellite data

The long-time technological trend towards ever-finer ground resolution in space-borne multispectral data has opened the doors to finer levels of urban mapping and monitoring. Single buildings and their features can nowadays be detected and mapped starting from nadiral data; yet, despite a large body of research results, an exhaustive solution to space-based building mapping is still to be found. In this paper, we give our contribution by proposing a novel approach to the extraction of rooftop shapes of buildings from very-high-resolution (VHR) optical multi-spectral data. The approach is derived from existing work, namely an automatic rooftop extraction method intended for aerial imagery. Because of the very different nature of the data, it was necessary to rearrange the reference method, modifying and adding new constraints, applying both spectral and spatial conditions. This work was developed in the framework of the EU H2020 Satellite Swarm Sensor Network (S3NET) project.