A geometry and texture coupled flexible generalization of urban building models

Abstract In the past, numerous research efforts have focused on generalization of city building models. However, a generic procedure for creating flexible generalization results supporting the fast and efficient update of original building models with various complexities is still an open problem. Moreover, building clusters created in previously published generalization methods are not flexible enough to meet the various requirements for both legible and realistic visualization. Motivated by these observations, this paper proposes a new method for generating a flexible generalization outcome which enables convenient updating of original building models. It also proposes a flexible preprocessing of this generalized information to render a legible and realistic urban scene. This is accomplished by introducing a novel component structure, termed as FEdge, particularly designed for efficiently managing the geometry and texture information in building cluster instances (both original building models and building clusters) during the generalization, visualization and updating processes. Furthermore, a multiple representation structure, referred to as Evolved Buffer-Tree (EBT), is also introduced. The purpose of the EBT is to organize building cluster instances and to employ more flexible LODs for both legible and realistic visualization of urban scenes. FEdge has an intuitive planar shape which can be effectively used in representing rough 3D facade composed by detailed continuous meshes. Each FEdge is given a unique identifier, referred to as FEdge Index. In the proposed generalization scheme, firstly each original building model treated as a building cluster instance is abstracted and presented as FEdge Indices. These FEdge Indices are then used for producing generalized building cluster instances in the EBT portably, and to support convenient model updating and flexible preprocessing of the generalization results for renderable building cluster instances. Secondly, to achieve a legible and realistic visualization of urban scene, the EBT is flexibly assigned diverse LODs maintaining more important legible information than LODs defined in CityGML for 3D building models. To make the generalization more accurate by considering the city roads and districts, an algorithm for automatic road analysis is applied in our clustering and combination. Numerous experiments considering the geometrical and textural complexity of common urban building models, as well as a typical case study of complex city scene with a large number of building models, verify the effectiveness of our generalization method and the dynamic visualization of the generalized urban models.