Automating 3D Building Modeling: A Comparative Study of Data- and Model-Driven Methods

Abstract. Accurate and efficient 3D building reconstruction from airborne point clouds is crucial for a wide range of applications, from urban planning to navigation and microscale simulations. This paper review and evaluate model- and data-driven approaches for building modeling, using two baselines’ methods and introducing two new implementations: an augmented, data‐driven variant of Kinetic Shape Reconstruction (KSR) and a model-driven approach under development named RoofGenerator. Experiments use a dataset available among the open-data of SwissTopo, featuring diverse building types and varying slopes, with manually reconstructed reference 3D building models providing a reliable ground truth for evaluation. The results show that data-driven methods offer greater flexibility and capture finer details but require careful parameter tuning and are sensitive to point density and segmentation accuracy. In contrast, model-driven approaches are computationally efficient and robust but constrained by predefined shape libraries, limiting their ability to model specific roof structures. Additional challenges include reconstruction consistency, footprint dependency and the lack of standardized evaluation metrics. The choice of the evaluation method depends on specific application needs, data quality and scalability requirements.