skip to content

Detection of walls, floors and ceilings in point cloud data (Automated As-Built Modelling)

Ioannis Anagnostopoulos, PhD candidate, and Ioannis Brilakis, Laing O' Rourke lecturer, research the detection and classification of objects like floors, walls and ceilings in point cloud data for automated as-built modelling. The creation of an as-is model is a complex and tedious procedure. It starts with laser scanning the facility to acquire point cloud data, a set of x,y,z coordinates. This point cloud data are raw without any semantic or useful information. Their modelling is done manually by experienced personnel, a process that requires extensive amounts of time and effort

The scope of this research is Manhattan-World (MW) buildings with extensive levels of clutter. The defining characteristic of a MW structure is that there are three mutually orthogonal directions. Hence, the present objects have distinguishing rules. For example, walls are parallel to each other and perpendicular to the floor and ceiling, whereas floors and ceilings are parallel to each other and parallel to the x-y plane.

The algorithm necessitates minimum human intervention, achieving an automated process for the detection of coarse objects in the point cloud. It achieves detection of multiple objects in a cluttered environment of multiple rooms, not only in isolated ones. The simplicity of the algorithm and the low computational complexity are additional contributions. 

The research starts with the segmentation of the point cloud into planar clusters. The floors and ceilings are separated by projecting the point cloud into the y-z and x-z plane, an octree division is applied and the maximum leap in point density is identified. The corresponding points in these leaps are the floors and ceilings. Walls are detected by exploiting the height, width and parallelism of the planar clusters. The interior walls are a pair of planar clusters, whereas the boundary walls are just one planar surface.

The proposed algorithm was tested on an original point cloud of an office building. The point cloud was registered and aligned with the appropriate axes. Furthermore, it was downsampled to ensure fast execution. The final result had a precision of 86.7% for exterior walls, 92% for interior walls and 100% for floors and ceilings.

The research leading to these results has received funding from the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreements n°247586 ("BIMAutoGen") and n°334241 ("INFRASTRUCTUREMODELS"). This Publication reflects only the author's views and the European Community is not liable for any use that may be made of the information contained herein.