M. Kada

University of Stuttgart



The article presents an automatic generalization approach for 3d building models with regard to the cartographic visualization of urban landscapes. Important application areas are urban planning, city marketing, location-based services and navigation. These applications generally do not run on dedicated graphics workstations, but rather on commodity PCs or even mobile devices like PDAs or cell phones. Here, the use of generalized building models speed up the rendering process and also improve the visual impression and perceptibility especially on small displays.

In order to reduce the complexity of building models, traditional mesh simplification techniques known from the field of computer graphics are not applicable. These algorithms are designed to reduce the complexity of arbitrarily shaped meshes that can consist of millions of triangles. The single buildings in a reconstructed 3d city model, however, typically feature a much smaller number of polygons. Furthermore, the regularities (e.g. rectangularity, coplanarity and parallelism) that are inherent to buildings must be preserved during simplification. Generalization need not only be limited to single buildings, but it can also be the merging of buildings in order to have a single model that forms a building block.

Based on a 3d building generalization technique that uses cell decomposition, a new approach is presented for the simplification of elements that feature curved ground plans. Generalization with cell decomposition first creates a decomposition of the ground plan from approximating planes. Since these planes can not maintain the features of curved elements, special care has to be taken. So before the actual generalization takes place, the line elements that approximate curved elements are detected in the ground plan polygon. This is e.g. the case for the towers of churches and castles which are important structures for these types of buildings. As a result, additional curved cells are generated for the aforementioned features.


Once the horizontal extension of the structure is known, then the roof type can be evaluated both for the rectangular and the curved cells. Here, an explicit rule set helps to ensure that neighbouring primitives fit together. From the roof types, parameterized primitives are computed that describe the building parts of each cell. For simplification purposes, the parameters for shed, saw tooth and parallel roof types can be altered in order to create fewer roof elements. This results in a typification of the roof. But also the ground plan of curved cells can be simplified by using fewer line segments. In order to gain the resulting 3d building model, the cells are combined with a combined gluing and Boolean intersection operation that are known from cell decomposition and constructive solid geometry.