A new benchmark on LoD 2 building reconstruction from aerial lidar and footprints

Abstract. The CityGML norm proposes specifications for the 3D representation of most urban objects, and in particular defines levels of details (LoD) for building models. The LoD2 corresponds to polyhedral roof structures representing the main roof slopes, but not superstructures as dormers and chimneys. Nowadays, several countries have a nationwide lidar program with densities well adapted to LoD2 building modeling and open topographic databases containing the building footprints, so automating LoD2 reconstruction from lidar and footprints allows the production of nationwide LoD2 models. This paper proposes a new benchmark to evaluate the quality, scalability and robustness of state of the art LoD2 building reconstruction from lidar and footprints. It is based on a subset of LiDAR HD data freely provided by IGN, the French mapping agency, along with building footprints derived from a high-quality, manually produced ground-truth dataset created by IGN. Results from various state-of-the-art algorithms are evaluated using the open source PyScoring tool that compares the results with the ground truth.