Insofar as a computer simulation is concerned, where there’s smoke, there’s not a fire but a large computational burden. Rendering smoky, foggy or cloudy scenes is difficult because the process requires substantial computational resources, time or both to get the view to closely match reality. Fog and smoke present such a problem because, unlike clear air or a vacuum, they are not passive when it comes to light transiting through them. In fact, they are known as participating media because, by absorbing or scattering passing light, they take an active role in determining how scenes look.A new computational method makes it more practical to re-create lighting within hazy scenes, such as this one of two cars on a foggy road. Images courtesy of Wojciech Jarosz, University of California, San Diego.Now a team from the University of California, San Diego, has developed a method that lightens the computational burden substantially with an algorithm that solves the math at only a sparse set of points, then uses the resulting values repeatedly at arbitrary locations. This not only lessens the drain on computers but also could result in more realistic looking video games and movies.There are a number of rendering solutions for such situations, with the most efficient, in general, being a technique called photon mapping. In this method, virtual sources spit out photons, and the path of each is computed and tracked. Although providing a significant performance improvement over brute force methods, photon mapping can be wasteful in some situations. For example, photons may shoot into parts of a scene never visible to the viewer, a possibility that is unanticipated when the generated virtual photons begin their trek. Consequently, photon mapping has trouble with large scenes or with scenes in which the camera or viewer is looking at only a small portion of the overall picture. On the left is a contrast-enhanced rendering of a scene created with a novel method called radiance caching, while on the right is the same scene made by a photon-mapping technique, currently used for such computerized renderings. The new method produces smoother results in the same amount of time.In their work, the researchers took a different approach, developing a method that concentrates effort where it is most important to a chosen viewpoint. Extending a technique that had been successfully used before, they created radiance caching, which resolves the lighting patterns at a sparse set of points, and then exploited the smooth nature of scattering in participating media to arrive at a close enough solution for an arbitrary point.In essence, this approach expands a point solution into a bubble throughout which the solution is valid enough. The researchers used a custom error-measuring metric to determine how big the bubble could be around a given point before the error grew too great and the rendering suffered too much. As project leader Wojciech Jarosz explained, “The error metric defines the density of cache points in different regions of the medium.”He noted that the technique adds cache points in a lazy fashion. For the lighting value at the very first point, the full value is computed along with a valid radius. Any subsequent computation involves first checking to see whether the point lies within any existing bubble, in which case the cached value is used. If not, further computation determines the lighting value and the valid radius, with the new results cached for later reuse.Tests indicated that the researchers’ approach is more efficient than photon mapping for certain scenes — a result that is a consequence of the new method’s being driven by the viewpoint of the observer, as they had hoped. However, Jarosz noted that photon mapping is better for simulating effects that are easier to detect when considered from the point of view of the light source. An example is the focusing of light by glass or mirrors, an effect known as volumetric caustics.The researchers hope to combine radiance caching and photon mapping in the future.SIGGraph 2007, Aug. 5 to 9, 2007, Jarosz et al, “Radiance Caching for Participating Media.”