超大规模地形场景的高性能漫游

High Performance Navigation of Very Large-Scale Terrain Environment

超大规模地形场景包含大量的几何和纹理数据,无法一次性载入内存,并具有极高的复杂度,因而无法进行实时绘制.提出一种高性能的外存地形场景实时漫游技术.该方法使用离散层次细节技术并结合视点相关的动态连续层次细节选择和过渡.算法为地表的简化提出一种新的基于受限法向锥的误差计算方法,使得模型简化具有轮廓保持和光照保持特性.当生成网格包含三角形数目相当时,该方法比传统的基于几何误差的简化更加符合漫游时视觉的感知规律.场景简化过程中提取出的潜在轮廓特征可以通过巧妙地构建漫游时视线方向上的增量地平线来随时更新场景不同部分的可见性信息,并以此控制无用数据页面的载入和无效场景部分的绘制,提高绘制速度.漫游系统采用多线程技术,使CPU,GPU,I/O三者的效率得到充分发挥,并可实时生成具有光照和阴影效果的漫游图像.

Interactive navigation of very large-scale terrain environment involves many difficulties because of its out-of-core geometric data and texture, which can not be wholly loaded into the internal memory. This paper puts forward a high performance technique for real-time walkthrough of the large-scale terrain environments. It performs geometric simplification within each terrain chunk during preprocessing and then generates a view-dependent continuous LOD of the whole scene with geomorphing at rnntime. Both silhouette preserving and shading preserving criteria are satisfied by applying a new error metric-constrained normal cone to view-independent simplification. The generated models for rendering is more suitable to human＇s perception than those by traditional geometric simplification criterion. Furthermore, the pre-extracted potential silhouette of each chunk can be u sed to construct the incremental horizon dynamically. This can prevent the ineffectual data from loading into the main memory and reduce the out-of-core paging and the number of triangles feeding to the renderer, thus lead to significant saving in interactive rendering. This paper also makes use of the advantage of GPU to generate lighting and shadow for large-scale outdoor environment, and balances the tradeoff between GPU, CPU and I/O through multi-thread.