3D building roof reconstruction from point clouds via generative models

This paper presents a generative statistical approach to 3D building roof reconstruction from airborne laser scanning point clouds. In previous works bottom-up methods, e.g., points clustering, plane detection, and contour extraction, are widely used. Since the laser scanning data of urban scenes often contain extra structures and artefacts due to tree clutter, reflection from windows, water features, etc., bottom-up reconstructions may result in a number of incomplete or irregular roof parts. We propose a new top-down statistical method for roof reconstruction, in which the bottom-up efforts mentioned above are no more required. Based on a predefined primitive library we conduct a generative modeling to construct the target roof that fit the data. Allowing overlapping, primitives are assembled and, if necessary, merged to present the entire roof. The selection of roof primitives, as well as the sampling of their parameters, is driven by the Reversible Jump Markov Chain Monte Carlo technique. Experiments are performed on both low-resolution (1m) and high-resolution (0.18m) data-sets. For high-resolution data we also show the possibility to reconstruct smaller roof features, such as chimneys and dormers. The results show robustness despite the clutter and flaws in the data points and plausibility in reconstruction.