Towards Complete LOD3 Models – Automatic Interpretation of Building Structures

The paper presents our work on the automatic reconstruction of complete LOD3 building models. Existing LOD2 models are refined by adding 3D façade geometry with explicit semantic information to the planar facades. Aiming at robust and realistic façade structures, independent of the quality and availabilty of sensor data, object knowledge in the form of a grammar is automatically inferred and applied. The grammar-based modeling concept is integrated in a learning system environment. Within a data-driven algorithm, basic façade elements such as windows and doors are automatically extracted and modeled from terrestrial LiDAR point clouds. After this interpretation step, superior structures and hierarchical relationships are derived from the obtained 3D façade model and transferred to a grammar-based description. This is implemented by means of a specially developed formal grammar, called façade grammar, which is appropriate to express the architectural style of a building through individually adapted rules. By applying such a façade grammar realistic façade structures can be generated even for building regions for which no sensor data or only measurements of poor quality are available. However, opportunities for grammar-based interpretation and modeling of building structures do not end at this point. The concept of describing geometrical and semantic relations in a formal way can also be applied for the representation of indoor environments as well as large-scale relationships between whole buildings or districts of a city. The final part of the paper proposes how the façade grammar could be integrated in a more general graph-oriented structure which allows for grammar-based modeling of buildings in various levels of details and different contextual environments.