LOD methods of large-scale urban building models by GPU accelerating

Abstract

The characteristics of rapid parallel computing of modern computer hardware are utilized in this paper. Hierarchy structures of CityGML building models are constructedin CPU dynamically by taking advantage of their spatial patterns and semantic relations. We present an approach for sending modification data to GPU buffers in batches so that communication between CPU and GPU can be reduced as much as possible. According to the data created by the hierarchy information dynamically in GPU, the corresponding level detail building models will be rendered under the using the properties of parallel handling of geometry shader. Therefore, powerful multi-threaded computing ability of GPU is utilized more effectively. The experiment results prove that our methods run more efficiently while consume less CPU and GPU footprints, and also can meet the demand of large-scale urban building models.