摘要
随着图形硬件性能的提升,大规模地形绘制的主要瓶颈已从绘制能力不足转变为大数据的传输,针对这一问题提出一种支持大规模地形的瓦片调度与实时绘制算法。将超大地形数据以瓦片金字塔形式存储于硬盘,绘制每一帧时只调度当前场景所需的少量瓦片进入显存。首先利用GPU实时计算地形网格点的地理坐标并传回CPU分析可见范围,然后采取瓦片四叉剖分、规则化处理和瓦片合并等一系列操作在所有LOD层中拣选最优瓦片集合并调入内存,在内存中利用一块固定大小的缓存进行管理与更新,并最终以单张纹理的形式传入显存进行采样和绘制。实验表明,该算法节约了大量的显存带宽,有效提升了系统在数据传输方面的执行效率,在大规模地形调度与绘制中取得了较好的效果。
Along with the development of the graphics hardware, the transmission of large-scale data has replaced the lack of rendering capacity as the new bottleneck of large scale terrain rendering. Aiming at this problem, we present a new approach for dynamic scheduling and real-time rendering for large-scale terrain. Terrain data are stored as tile-pyramid in hard disk, in real-time rendering phase, only a few of tiles are streamed into the video memory every frame. First, GPU computes and transfers geographic coordinates of the current scene into CPU in real time,CPU analyses and calcu- lates the visible region through these coordinates, and then selects the optimum combination of visible tiles（visible tiles set） among all the layers by quad-tree generating, tiles regularization and tiles merging. The visible tiles set are streamed into CPU, managed and updated in a fix-sized buffer, and finally transferred into GPU as a texture for sampling and rendering. Experiments show that this algorithm saves a large amount of video memory bandwidth, improves the efficiency of data transmission effectively,and obtaines a good result in the large scale terrain scheduling and rendering.
出处
《计算机科学》
CSCD
北大核心
2013年第06A期120-124,共5页
Computer Science
基金
国家自然科学基金项目(41001300)资助
