Ambient Occlusion

AO OffAO On

大脑对光影变化很敏感，因此可以通过光影构造空间理解。

由于附近几何的遮挡，有些点朝某些方向的光被挡住，产生了光影的效果。

Precomputed AO

Using ray tracing to compute the AO offline and store the result into texture, which is widely used in object modeling process

  • Extra storage cost
  • Only apply to static object

对于有特点的角色，可以预计算 AO

Screen Space Ambient Occlusion (SSAO)

对于一个场景，场景中对象的位置不固定，不能预计算。利用一个点周边的采样情况来估算它的被遮挡情况。

  • Generate N random samples in a sphere around each pixel p in view space

  • Test sample occlusions by comparing depth against depth buffer

  • Average visibility of sample points to approximate AO

$$ A(p)=1-\frac{Occlusion}{N} $$

公式不重要，重要的是思想：局部采样。

SSAO+

改进版：只采样半球。

  • Recall the AO equation is acutally done on the normal-oriented hemisphere

HBAO - Horizon-based Ambient Occlusion

采样改成积分。

  • Use the depth buffer as a heightfield on 2D surface
  • Rays that below the horizon angle are occluded

先计算一圈每个方向的仰角，可以估算出有多大面积的天顶是可见的。
Attenuation Trik：如果遮挡物离当前点比较远，就不产生影响。

  • Trace rays directly in 2D and approximate AO from horizon angle

实际算法中有很多细节。

GTAO - Ground Truth-based Ambient Occlusion

贡献1

HBAO 没有考虑到角度的因素。因为从不同角度射下来的光，对点的影响是不一样的。

GTAO introduces the missing cosine factor, removes the attenuation function, and add a fast approximation of multi bounce

贡献2

Add multiple bounces by fitting a cubic polynomial per albedo

用真实数据拟合了 Single Bounce AO 与 Multi Bounce AO 的关系(假设为三阶多项式)

Result

Ray-Tracing Ambient Occlusion

  • Casting rays from each screen pixel using RTT hardware
    • 1 spp(sample per-pixel) works well for far-field occlusion
    • With 2-4 spp, can recover detailed occlusion in contact region

(1) 利用 GPU 的 ray casting 能力
(2) 每帧只 sample 一个方向，在时序上持续收集遮挡信息。

本文出自CaterpillarStudyGroup，转载请注明出处。
https://caterpillarstudygroup.github.io/GAMES101_mdbook/