Generative Modelle für die Erfassung und Generalisierung von Stadtmodellen
Final Report Abstract
In the research conducted in this project we have proposed and implemented a generative statistical approach for the extraction and reconstruction of building roofs from airborne laser scanning point clouds. The main contributions of this work can be summarized as follows: A novel primitive-based modeling scheme with combination rules specially designed allowing primitive overlapping; A generative statistical reconstruction driven by reversible jump MCMC with explicit model selection methods; Extension of the proposed modeling scheme for the building model generalization. Additionally, for practical applications we conducted a pre-segmentation procedure dealing with large and complex urban scenes. By all these means an automatic building roof extraction framework is presented with the robustness against data flaws and clutter objects and the plausibility of reconstruction. Although we have shown the power and flexibility of the top-down methods, they have their own issues of uncertainty and instability. Generative modeling is driven by the random sampling and the runtime and final results could be significantly influenced by the prior knowledge (initial state, parameter priors) and the scene complexity. E.g., large and sophisticated city blocks, which cannot be divided into smaller parts by the pre-segmentation, may lead to incomplete and varied result after each run of the program. Concerning possible future work we, therefore, first consider integrating more bottom-up information. Cadastral maps, for instance, will provide reliable initial information for individual buildings and help to divide the point cloud more reasonably. Some cadastral maps have additional building height information, which can be used as priors for ridge and eave heights. By all these means the reconstruction results will become more stable and the computational effort can be reduced. In this work our primitive library contains only planar roofs with at most three differing height levels. New entries, e.g., flat roofs in the forms of triangle and ellipse, domes, cones, and other curved shapes, are needed to present more sophisticated buildings like churches, exhibition centers and stadia.
Publications
- 2008. Building reconstruction using a structural description based on a formal grammar. International archives of photogrammetry, remote sensing and spatial information sciences, vol. XXXVII, Beijing, China
Milde, J., Zhang, Y., Brenner, C., Plümer, L. and Sester, M.
- 2009. Graph-based modeling of building roofs. In: Proceedings of 12th AGILE conference on GIScience, Hannover, Germany
Milde, J. and Brenner, C.
- 2011. 3D building roof reconstruction from point clouds via generative models. In: Proceedings of 19th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (GIS), Chicago, IL, USA, pp. 16–24
Huang, H., Brenner, C. and Sester, M.
- 2011. A hybrid approach to extraction and refinement of building footprints from airborne LIDAR Data. In: Proceedings of ISPRS Workshop on Geospatial Data Infrastructure: from data acquisition and updating to smarter services, October, Guilin, China, pp. 153–158
Huang, H. & Sester, M.
- 2011. Rule-based roof plane detection and segmentation from laser point clouds. In: Proceedings of Joint Urban Remote Sensing Event (JURSE), April, Munich, Germany, pp. 293–296
Huang, H. & Brenner, C.
