Flying Voxel Method with Delaunay Triangulation Criterion for Façade/Feature Detection for Computation

Abstract

ABSTRACT:In an effort to reconstruct geometric models of building façades from terrestrial laser scanning data directly without either manual intervention or any third party computer-aid design package, a new algorithm is introduced. The algorithm detects building boundaries and features and converts the point cloud data into a solid model appropriate for computational modeling. The algorithm combines a voxel-based technique with a Delaunay triangulation based criterion. In the first phase, the algorithm detects façade boundary points from raw data. The algorithm's second phase creates a solid model using voxels in a quadtree representation.Finally, the algorithm determines whether holes are actual openings or data deficits caused by occlusions and then fills unrealistic openings. The algorithm was applied to the façades of three masonry buildings. For these buildings, the algorithm successfully detected all openings and reconstructs the façade details correctly. Geometric validation of the models against measured drawings showed overall dimensions correct to 1.2%, mostly opening areas to 3%, and simulation results within 5% of those predicted by a CAD-based model.

Octree-based, automatic building façade generation from LiDAR data

“…The geometric building model generated from this algorithm is subsequently imported directly into a finite element package for structural analysis. To achieve this, both topology and geometry of the building model are converted into suitable data formats by a Boundary-Representation (BRep) [6]. Sample points are defined as a set of point, P i (i = 1,…,n), which have three coordinates, (x i , y i , z i ) R 3 in Cartesian coordinate system.…”

“…doors or windows), if their characteristics satisfy those of common openings in building façades involving minimum lengths and widths equal or greater than 0.4m [6,7,12,21] and a height-to-length ratio between 0.25 and 5.0 [38,41] …”

“…Finally, the boundary lines are determined from the boundary points within the full voxels of these voxel sub-clusters by using a least-squares method (Figure 10b) [42]. Additionally, the boundary lines of each building feature are adjusted to ensure continuity by extending or trimming the boundary lines at their points of intersection (Figure 9c Subsequently, each voxel in the re-voxelization model is characterized by using the Flying Voxel method proposed by Truong-Hong et al [6], in which voxels inside of openings or outside of the façade are labelled either as empty or as full. Figure 9e shows the recharacterized voxels of the octree voxelization in Figure 9d.…”

“…Figure 9e shows the recharacterized voxels of the octree voxelization in Figure 9d. Subsequently, the full voxels in the octree nodes are stored in a neutral file (with the file extension .ANF), whereas the topology and geometry are described as a B-Rep scheme [6].…”

“…Laser scanning is gaining popularity in reverse engineering [1][2][3] and visualization [4,5], yet exploiting the data for structural analysis remains a topic for development [6,7]. Developing automatic, robust, and efficient methods to reconstruct geometric building models for computational modelling has the potential to save money and time for projects needing cityscale modelling.…”