Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

A solution to global illumination by genetic algorithms

: Hornung, C.; Lange, B.

Albrecht, R.F.; Reeves, C.R.; Steele, N.C.:
Artificial neural nets and genetic algorithms. Proceedings of the International Conference 1993
Wien: Springer, 1993
ISBN: 0-387-82459-6
ISBN: 3-211-82459-6
International Conference on Artificial Neural Nets and Genetic Algorithms (ANNGA) <1, 1993, Innsbruck>
Fraunhofer IGD ()
crowding; genetic algorithms; global illumination; Monte Carlo methods; sharing; stochastic ray-tracing

A new approach to optimize the computer simulation of radiant light transfer by means of evolutionary techniques for the generation of photorealistic images is introduced. The formulation of radiant light transfer in a model leads to a system of complex integral equations, which currently have been solved by Monte Carlo Methods. One of the major problems in Monte Carlo sampling is to determine the location and density of sample points in order to reduce the variance of the estimates. Here a solution is provided by applying evolution strategies to calculate the global illumination. Thus exploiting the search space, i.e. the hemisphere of incident radiation to a point on a surface in a very efficient way through maintaining populations of rays and applying selfadaptive genetic recombination operators. The simulation process now becomes selforganizing and the transition of one state into another is no longer independent of previous states which allows the system to adjust optimally to a p articular lighting situation.