Automatic Generation of Structural Building Descriptions from 3D Point Cloud Scans
Abstract
We present a new method for automatic semantic structuring of 3D point clouds representing buildings. In contrast to existing approaches which either target the outside appearance like the facade structure or rather low-level geometric structures, we focus on the building's interior using indoor scans to derive high-level architectural entities like rooms and doors. Starting with a registered 3D point cloud, we probabilistically model the affiliation of each measured point to a certain room in the building. We solve the resulting clustering problem using an iterative algorithm that relies on the estimated visibilities between any two locations within the point cloud. With the segmentation into rooms at hand, we subsequently determine the locations and extents of doors between adjacent rooms. In our experiments, we demonstrate the feasibility of our method by applying it to synthetic as well as to real-world data.
Images
![]() |
Download Paper
Additional Material
- Presentation slides (PDF document, 48 MB)
Bibtex
@INPROCEEDINGS{ochmann-2014, author = {Ochmann, Sebastian and Vock, Richard and Wessel, Raoul and Tamke, Martin and Klein, Reinhard}, title = {Automatic Generation of Structural Building Descriptions from 3D Point Cloud Scans}, booktitle = {GRAPP 2014 - International Conference on Computer Graphics Theory and Applications}, year = {2014}, month = jan, publisher = {SCITEPRESS}, location = {Lisbon, Portugal}, abstract = {We present a new method for automatic semantic structuring of 3D point clouds representing buildings. In contrast to existing approaches which either target the outside appearance like the facade structure or rather low-level geometric structures, we focus on the building's interior using indoor scans to derive high-level architectural entities like rooms and doors. Starting with a registered 3D point cloud, we probabilistically model the affiliation of each measured point to a certain room in the building. We solve the resulting clustering problem using an iterative algorithm that relies on the estimated visibilities between any two locations within the point cloud. With the segmentation into rooms at hand, we subsequently determine the locations and extents of doors between adjacent rooms. In our experiments, we demonstrate the feasibility of our method by applying it to synthetic as well as to real-world data.} }