Research On GPU-Based Large-Scale Terrain Rendering

Large-scale terrain scene is an important part of virtual battlefield. Real-time interactive rendering of terrain scene is a great challenge to graphics hardware due to its massive datasets. So we have to study terrain simplification algorithms, which finally decide how much geometry will be sent to rendering pipeline, in order to improve the rendering frame rate.Terrain simplification algorithms invented in the past are not suitable to be used in GPU-Based graphics processing framework anymore. The algorithms should resort to more conservative simplification methods by pushing as much triangles through the pipeline as GPU can handle, with the least amount of CPU overhead.The thesis studies two kinds of terrain simplification algorithm, level of detail and visibility culling technique in conjunction with GPU characteristic respectively:1. The thesis introduces a GPU-Friendly Patch-LOD terrain rendering algorithm based on GeoMipMapping method: HIT, Hierarchical Index Templet. The algorithm simplifies and speeds up the creation of multi-resolution mesh for terrain by using an index buffer sequence, which resides in video memory, to represent terrain blocks' detail levels.2. The thesis solves the cracking and popping problem respectively on GPU based on the HIT algorithm. The crack between adjacent terrain blocks with different resolution is eliminated by extending the HIT with joint index templet. Moreover, Geo-Morph templet is introduced to implement Geo-Morphing in vertex shader stage of rendering pipeline.3. Hardware occlusion query is used to perform occlusion culling for terrain based on well-known occlusion horizon algorithm. In order to make use of hardware occlusion query, the thesis uses stencil buffer to perform the occlusion test. Moreover, query list is used to avoid the latency between CPU and GPU caused by issuing the query and availability of the result.A GPU-Based framework for large-scale terrain rendering is implemented using the above algorithms in the end. The experimental results demonstrate that the framework develops GPU performance, reduces CPU overhead and improves the rendering frame rate significantly. It is an easy, fast and flexible method for real-time interactive rendering of large-scale terrain scene.