Volume II-3/W3
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., II-3/W3, 73-78, 2013
https://doi.org/10.5194/isprsannals-II-3-W3-73-2013
© Author(s) 2013. This work is distributed under
the Creative Commons Attribution 3.0 License.
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., II-3/W3, 73-78, 2013
https://doi.org/10.5194/isprsannals-II-3-W3-73-2013
© Author(s) 2013. This work is distributed under
the Creative Commons Attribution 3.0 License.

  08 Oct 2013

08 Oct 2013

Roof Reconstruction from Point Clouds using Importance Sampling

W. Nguatem, M. Drauschke, and H. Mayer W. Nguatem et al.
  • University of Bundeswehr, Munich, Institute of Applied Computer Science, Chair for Visual Computing, Neubiberg, Germany

Keywords: Building Reconstruction, Point Cloud Segmentation, MCMC, Model Selection, Non-linear Regression Fitting

Abstract. We propose a novel fully automatic technique for roof fitting in 3D point clouds based on sequential importance sampling (SIS). Our approach makes no assumption of the nature (sparse, dense) or origin (LIDAR, image matching) of the point clouds and further distinguishes, automatically, between different basic roof types based on model selection. The algorithm comprises an inherent data parallelism, the lack of which has been a major drawback of most Monte Carlo schemes. A further speedup is achieved by applying a coarse to fine search within all probable roof configurations in the sample space of roofs. The robustness and effectiveness of our roof reconstruction algorithm is illustrated for point clouds of varying nature.