Stochastic Billboard Clouds for Interactive Foliage Rendering

Dylan Lacewell, Dave Edwards, Peter Shirley, William B. Thompson

School of Computing, University of Utah


We render tree foliage levels of detail (LODs) using a new adaptation of billboard clouds. Our contributions are a simple and efficient billboard cloud creation algorithm designed specifically for tree foliage, and a method for smooth transitions between LODs. The cloud creation algorithm performs stochastic search to find a set of billboards that approximate the base mesh. Billboard clouds offer an alternative to traditional triangle reduction methods, which break down for foliage with its small, disconnected pieces of geometry. By projecting foliage geometry onto large precomputed textures, our method shifts the bulk of the runtime rendering to the fragment processing stage. This results in higher framerates for most viewing distances, with adjustable visual accuracy. We give results for foliage from two fully detailed tree models and discuss implementation issues.

Update (July 2008): Some other researchers have also used billboard clouds for foliage rendering. We recently became aware of the paper below from 2005 which presents work similar to ours:

jdl 07/2008