To improve the real-time performance of computing approximate global illumination in virtual scenes, an adaptive refinement sampling algorithm is presented based on reflective shadow maps. A mulit-resolution sampling meth-od is proposed through analyzing the variance of the depth of a shadow map. Analyzes distance relationship between points with their world space coordinates in shadow maps and estimates the distribution of virtual point lights within sampling area to generate a multi-resolution sampling method and computes indirect illumination with these different sampling resolutions. Experiments show that the algorithm can reduce about 20% of the sample points in the best cases and keep the same qualities compared with the reflective shadow maps. Our method builds two databases for indirect il-lumination and sampling for GPU-CPU task division, computes and stores the proportion of depth, normal, flux, depth variance and sampling sub-levels in fragment shader separately. Compares different sort of task division based on the heterogeneous GPU-CPU computing platform for indirect illumination with virtual point lights to enforce a real ef-fective load balancing and peak operation.%为改进虚拟场景的近似全局光照计算的实时性，提出一种反射阴影图的采样自适应细化算法。通过阴影图深度方差分析，设计一种阴影图采样点的层次划分方法。分析采样点在世界空间坐标系下的距离关系并估算采样区域内有效虚拟点光源的分布，对阴影图进行层次划分。利用不同的采样分辨率计算间接光照。实验表明，自适应细化采样与反射阴影图的采样算法相比，在最好情况下，可以减少20%的采样数量，并保持画面质量。算法为CPU和GPU之间的任务划分建立两个间接光照及采样数据库，分别计算和存储场景点的深度、表面法向量、flux通量、方差及片段着色器中采样的分级结构，记录任务的划分比例，通过异构CPU-GPU平台对不同任务划分及虚拟点光源的间接光照计算对比，实现有效的负载平衡和运算峰值。
展开▼
