Preparing simplified 3D scenes of multiple LODs of buildings in urban areas based on a raster approach and information theory

Abstract

We have developed a method for the simplification of the footprints of 2D buildings based on a rasterisation process. The rasterisation is processed within quarters and the urban area is subdivided into quarters based on natural contours such as roads and water objects and not on straight geometric lines (the common subdivision approach to orthogonal tiles). Quarters were organised into a hierarchical model, according to the gaps between the quarters and the stages of the clustering process, using Kohonen’s self-organising maps. Each degree of simplification (generalisation) corresponds to some level of hierarchy. Information theory was used to estimate the amount of the 2D generalisation of building footprints. Simplified building footprints were extruded for the compilation of a 3D urban perspective from multiple levels of detail (LODs) The entropies of 3D scenes for each quarter of the hierarchy and each LOD were compared in order to define the level of detail to be used in the final 3D scene.