3D surface reconstruction

The surface reconstruction of buildings is the most crucial aspect in UBEM because a surface which is not airtight will lead the building energy simulation to fail. The reason is that EPlus cannot determine the volume of an open geometry and will stop with an error. In previous UBEM studies, the 3D model of a building was approximated by extruding its footprint. This approach leads to LoD1, where the roof is flat. In order to achieve an higher LoD with a fully formulated roof, the footprints are combined with a city-wide point cloud. The footprints of buildings can be retrieved from OpenStreetMap (OSM) [44]. The point cloud of a city is generated by LiDAR, which is a set of points defined by geographic coordinates (x, y) and height information (z) in Figure 2. It is normally used in the creation of Digital Elevation Maps. Figure 2 demonstrates how the footprint of a building is applied as a filter to selected the points constituting the roof section of a building. After isolating the point cloud of the roof, it has to be cleaned from outliers, such as trees and reflections from windows. The next step detects the heights of the walls. This requires the identification of the roof outline by associating the heights of the closest points to the edges of the footprint. The resampling of a point cloud from the entire building envelope, including walls and floor, finalizes the point cloud processing.
The screened Poisson surface reconstruction algorithm [32] is subsequently employed to generate a triangle mesh of the building. This algorithm ensures an airtight mesh, which is a collection of faces, edges and vertices defining shapes of three-dimensional objects. An alternative would be to merge the easily determined floor and wall surfaces with the triangular roof surfaces, but this approach is not trivial in three-dimensional space. Screened Poisson surface reconstruction is implemented in the 3D modelling tool Meshlab [8]. Besides mesh generation, Meshlab offers many other auxiliary algorithms and the convenience of a command-line interface to automate the process. The meshes are exported as Collada file type, which is a common file format for 3D applications. Figure 3 shows the results of the mesh generation and surface classification.

Figure 2: Point cloud (point colors refer to elevation) and OSM footprint of a building.

Figure 3: Three-dimensional model of a building with surface classification (roof in blue, walls in orange and floor in green).