Hybrid representations and perceptual metrics for scalable human simulation

Abstract

The simulation of large crowds of humans is important in many fields of computer graphics, including real-time applications such as games, as they can breathe life into otherwise static scenes and enhance believability. Such applications need to deal with having limited resources available at each frame. With many hundreds or thousands of potential virtual humans in a crowd, traditional techniques rapidly become overwhelmed and are not able to sustain an interactive frame-rate. Therefore, simpler approaches to the rendering, animation and behaviour control of the crowds are needed. Additionally, these new approaches must provide for variety, as environments inhabited by carbon-copy clones can be disconcerting and unrealistic.