A Terrain Rendering Algorithm Based On GPU Acceleration

Terrain visualization technique commonly used in the virtual environment, geographic information systems. In terrain visualization system, usually LOD (level of detail) technology is used to achieve view-dependent dynamic network surface construction. With the terrain sampling technology, more and more high-resolution topographic data, the amount of data is also growing, the traditional LOD algorithms faces in building a terrain mesh terrain data to read all of the White deposit, generate the terrain mesh surface then transferred to the card through the graphics library for rendering the terrain data is large, the data transmission efficiency will decline. In addition, in recent years with the rapid development of GPU technology, more and more people try to use only the CPU before the GPU to complete many tasks to be processed. Therefore, the paper focuses on LOD algorithm is implemented on the GPU problem, mainly the following:An adaptive algorithm converts terrain data into two-dimensional texture. LOD algorithm to build the traditional network of surface topography, the terrain data are used directly on top-down approach build network side, although relatively simple algorithm, but the terrain data is large, the top surface of the data network is the largest span This situation will lead to the operating system frequently read the disk, but the amount of data per read the disks are very small, greatly reduces the efficiency of the system. This paper presents an adaptive algorithm for terrain two-dimensional texture to the current parameters as the cone, GPU's video memory capacity for the input, output, sub-optimal number of blocks, then each sub-block building into a two-dimensional texture and texture Send to the memory, GPU video memory to read directly from the network of surface elevation values and build, greatly improved rendering speed.A GPU can operate in the organization of data tree structure (quad stack) and the corresponding LOD terrain rendering algorithms. As the GPU's instruction set does not contain a pointer operation, the traditional LOD algorithm uses quadtree, binary tree structure of another triangle contains a pointer field, in order to enable GPU to operate the hierarchical structure, this paper quad heap. The network structure of both the surface topography can be stored in the level of information, but also support an array of vector operations, realized in the implementation of LOD on GPU data structures.The design and implementation of a GPU-based terrain visualization system accelerated. In this paper, C++and OpenGL environment, with Nvidia Cg language based on quad heap LOD algorithm and experimental GPU's geometry shaders to run, experimental results show that rendering the same image, the algorithm is frame rate, triangle throughput and bandwidth are superior to the traditional four-tree algorithm.