Volume II-3/W4
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., II-3/W4, 17-24, 2015
https://doi.org/10.5194/isprsannals-II-3-W4-17-2015
© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., II-3/W4, 17-24, 2015
https://doi.org/10.5194/isprsannals-II-3-W4-17-2015
© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.

  11 Mar 2015

11 Mar 2015

GRAMMAR-SUPPORTED 3D INDOOR RECONSTRUCTION FROM POINT CLOUDS FOR “AS-BUILT” BIM

S. Becker, M. Peter, and D. Fritsch S. Becker et al.
  • Institute for Photogrammetry, University of Stuttgart, 70174 Stuttgart, Germany

Keywords: Building, Modeling, Abstraction, Prediction, Automation

Abstract. The paper presents a grammar-based approach for the robust automatic reconstruction of 3D interiors from raw point clouds. The core of the approach is a 3D indoor grammar which is an extension of our previously published grammar concept for the modeling of 2D floor plans. The grammar allows for the modeling of buildings whose horizontal, continuous floors are traversed by hallways providing access to the rooms as it is the case for most office buildings or public buildings like schools, hospitals or hotels. The grammar is designed in such way that it can be embedded in an iterative automatic learning process providing a seamless transition from LOD3 to LOD4 building models. Starting from an initial low-level grammar, automatically derived from the window representations of an available LOD3 building model, hypotheses about indoor geometries can be generated. The hypothesized indoor geometries are checked against observation data - here 3D point clouds - collected in the interior of the building. The verified and accepted geometries form the basis for an automatic update of the initial grammar. By this, the knowledge content of the initial grammar is enriched, leading to a grammar with increased quality. This higher-level grammar can then be applied to predict realistic geometries to building parts where only sparse observation data are available. Thus, our approach allows for the robust generation of complete 3D indoor models whose quality can be improved continuously as soon as new observation data are fed into the grammar-based reconstruction process. The feasibility of our approach is demonstrated based on a real-world example.