Realtime Ray Tracing and Interactive Global Illumination - Scientific ...

24 Chapter 3: A Brief Survey of Ray Tracing Acceleration Methods
As such, practical scenes (which usually contain both many light sources
as well as specular objects) require a large number of rays per pixel. Reducing
this number can then lead to significant improvements in rendering time.
3.2.1 Shadow Caching
Typically, the majority of all rays shot in a ray tracer are shadow rays,
as each primary or secondary ray usually triggers many shadow rays. For
shadow rays, however, it is sufficient to find any occluding object to guarantee
occlusion. Furthermore, many shadow rays are very similar (e.g. all shadow
rays to the same point light source), and are often occluded by the same
object. “Shadow caching” [Haines86] exploits this coherence by storing at
each light source the last occluder. Each new shadow ray to this light source is
then first intersected with this cached occluder. If any of these cached object
yields a positive intersection, occlusion of the shadow ray is guaranteed,
thus the shadow ray does not have to be traced any more. For many light
sources with a high degree of occlusion, shadow caching can greatly reduce
the number of shadow rays. Furthermore, shadow caching does not require
any approximations, does not generate any artifacts, and is orthogonal to
most other techniques (i.e. it can easily be combined with other techniques).
However, shadow caching also has some drawbacks: For example, it
quickly breaks down if the primary and shadow rays get incoherent, as
this leads to “thrashing” of the shadow caches. This is especially true for
area light sources and global illumination algorithms. Even worse, by design
shadow caches can only accelerate shadow rays that are actually occluded.
For non-occluded shadow rays, there is even some overhead due to the cost for
intersecting with the cached occluder. Finally, shadow caching often breaks
down in complex scenes, as small triangles are less likely to occlude several
successive shadow rays. Due to these reasons, shadow caching is generally
not used in the RTRT/OpenRT engine.
3.2.2 Local Illumination Environments
Recently, a more sophisticated optimization for shadow rays has been proposed
in the form of “Local Illumination Environments” [Fernandez02] (also
dubbed “LIEs”): LIEs subdivide the scene into a set of “voxels”, each of
which stores information on how the different light sources influence that
respective region of space. Each voxel stores a list of light sources that will
be fully occluded, fully visible, or partially visible with respect to that voxel.
With this information, the number of shadow rays to be show can be significantly
reduced: Light sources that are fully occluded from the respective