Techniques for generalizing building geometry of complex virtual 3D City models

Abstract

Comprehensible and effective visualization of complex virtual 3D city models requires an abstraction of city model components to provide different degrees of generalization. This paper discusses generalization techniques that achieve clustering, simplification, aggregation and accentuation of 3D building ensembles. In a preprocessing step, individual building models are clustered into cells defined by and derived from its surrounding infrastructure network such as streets and rivers. If the infrastructure network is organized hierarchically, the granularity of the cells can be varied correspondingly. Three fundamental approaches have been identified, implemented, and analyzed: The first technique uses cell generalization; from a given cell it extrudes a 3D block, whose height is calculated as the weighted average of the contained buildings; as optimization, outliers can be managed separately. The second technique is based on convex-hull generalization, which approximates the contained buildings by creating the convex hull for the building ensemble. The third technique relies on voxelization, which converts the buildings' geometry into a regular 3D raster data representation. Through morphological operations and Gaussian blurring, aggregation and simplification is yielded; polygonal geometry is created through a marching cubes algorithm. The paper closes with conclusions drawn with respect to the characteristics and applicability of the presented generalization techniques for interactive 3D systems based on complex virtual 3D city models.