Generating Method Based On GPU Acceleration For Fractal Terrain

Terrain plays an important role in virtual reality such as the simulation of natural phenomenon and vision system. Now, the generating method based on Digital Evaluation Model is used more frequently among all the methods, but it needs heavy calculation and high complexity. Real-time simulation is expected in virtual reality and so the efficiency of algorithm is more important. The fractal approach is widely used because of its briefness and high performance and gets nice results. However, this method is not as effective as expected when generating large scale terrain because of heavy calculation.Therefore, this paper raises a new generating method based on GPU acceleration for fractal terrain. This method converts the Diamond-Square algorithm into the process of parallel CUDA thread blocks, and then the generation for terrain is executed completely through GPU. So the algorithm increases speed evidently and meanwhile it achieves a good effect.Firstly, the development of 3D fractal terrain and the theory of fractal theory are studied in this thesis. Also, some typical algorithms of terrain generating such as DEM method, fractal method and data fitting method are compared in this thesis.Secondly, introducing the development of GPU and its architecture and NVIDIA CUDA. GPU has been widely used in Computer Graphics because of its parallelism.Thirdly, due to the existing problems such as heavy calculation and low speed, this paper raise a generating method based on GPU acceleration for fractal terrain. Taking advantage of GPU parallelism, this method converts the recursive or iterated Diamond-Square algorithm into concurrent computation, then the algorithm increases speed evidently. Meanwhile, the comparison of CPU and GPU time consuming is showed in this thesis. The result prove that the new algorithmn is very effective in generating large scale terrain.Finally, in order to get more vivid terrain, several terrain rendering methods are discussed and applied. The effect diagram is given in this paper.