Aggregation of 3D Buildings Using a Hybrid Data Approach
Kada, Martin, Cartography and Geographic Information Science
The adequate presentation of 3D landscape models--particularly of urban regions--becomes increasingly important as the availability of such data sets increases. Although it has already been demonstrated that the real-time rendering of completely textured 3D city models is possible with current commodity 3D graphics hardware (Kada et al. 2003; Buchholz and Dollner 2005), different levels of detail are still essential for network transmission visibility, analysis and visualization, especially for mobile applications like pedestrian navigation systems. As photorealistic views pose huge information loads on humans, the 3D building models need to be generalized under cartographic considerations with the purpose to make spatial situations easier and faster to comprehend.
While there has been and continues to be much work done in the 2D domain (Haunert and Wolff 2010), the topic of 3D generalization is rather new. Only few approaches have so far been presented, and most focus on the geometric simplification of single buildings. Whereas surface simplification approaches known from the field of computer graphics are meant for general shapes, the simplification of 3D building models in a cartographic sense is commonly expected to preserve existing shape symmetries and regularities like co-planar, parallel, and rectangular alignment of facade walls. For this purpose, 3D generalization approaches have been proposed that add restrictions to surface simplification operators (Coors 2001; Kada 2002; Rau et al. 2006), use mathematical morphology and specially designed curvature-space operators (Forberg 2004), apply feature segmentation, recognition, and elimination (Thiemann and Sester 2004), match and replace complex shapes with coarse parameterized templates (Thiemann and Sester 2006), and detect shape symmetries by crystallographic analyses (Poupeau and Ruas 2007).
However, for 3D city models that are currently being produced for large areas, their geometric complexity does not come from the individual buildings, as the majority thereof are of rather low detail, but from the large number of entities that can easily reach into the hundreds of thousands. Only few landmarks are modeled by means of terrestrial data, or by hand feature high detail. Therefore, the simplification process on its own has a natural limit up to which the complexity of such models can be reduced without violating the above-mentioned restrictions. For example, a single-family house with a saddleback roof can be simplified to a rectangular ground plan with four facade sides and two roof faces; maybe one if the roof is generalized to a flat roof. But a further simplification is not possible without losing the characteristics of a building or deleting it entirely from the data set.
The classic motivation of cartographic generalization states that if a feature cannot be properly presented in a map due to its scale, then it should be eliminated, accentuated, or replaced by something that allows for better readability Regarding closely located buildings that are often found in urban areas that cannot be distinguished as single objects, their aggregation to a larger building block can help to further reduce the complexity without changing the appearance of the spatial situation. Small gaps or even wider openings between the buildings are eliminated in the process with regard to some distance criteria. One of the few approaches for the geometric aggregation of 3D buildings is presented in Anders (2005): building groups are projected into three orthogonal directions, their silhouettes simplified in 2D with the ground plan generalization software CHANGE, the results extruded and then intersected to form the generalized 3D building group. For the purpose of interactive visualizations of 3D city models, Glander and Dollner (2009) show abstraction techniques that also include the aggregation of city blocks. Depending on the context, important landmarks are maintained in their original state to better help the viewer to focus on the current task. …