Visual Behavior Definition for 3D Crowd Animation through Neuro-evolution

This paper addresses the problem of creating crowd based scenes in animated films automatically. The main problem in this area is how to provide a natural way for the animator or director to define what they want the crowd to do. To this end, we propose here a hybrid neuro-evolutionary scheme where the artists regulate the behavior that is desired from the crowd by drawing colored lines and areas within a scenario. These elements are then transformed into energy based aggregative fitness functions that can be used to evaluate the behaviors of the individuals within the crowd during the evolutionary process that produces the controller for all the characters and, consequently, determines the behavior of the crowd as a whole. The approach has been tested on several different real scenes within the workflow of a local animation film company and the results it produced were very satisfactory.