Real-time rendering of large-scale terrain based on GPU

Abstract

For efficient visualization, storage and management of massive datasets in current large-scale terrain rendering system, a real-time rendering algorithm of large-scale terrain based on GPU (graphic processing unit) is proposed. The algorithm is an out-of-core supported aggregated LOD (level of detail) rendering method and focuses on the use of GPUs' ability in the computation and programmability. The two-ratio filtering simplification method and quadtree-based chunk hierarchy are used to generate a simple and regular chunk structure for view-dependent LOD visualization. Based on this structure, a GPU-oriented data compression strategy is given for reducing data transfer bandwidth and storage consumption; a template-based triangle stripification method is used for optimizing visualization efficiency and generating a conforming terrain without cracks. Besides the management strategy of massive datasets which is based on a two-level caching mechanism makes the algorithm breakthrough the hardware limitation and realize the out-of-core rendering. During the typical fly-over for large-scale datasets, a very high frame rate can be guaranteed by the algorithm.