Level of detail representations and variation methods for the rendering of large animated crowds

Abstract

The simulation of large crowds of characters is an important topic in the field of real time rendering as games and movies strive for more realistic populated scenes. Displaying a city scene with a large crowd enhances the believability of a movie and immersive feeling of a game. What movies and games have in common is that the assets used to create the crowd, such as models, textures and animations must be generated by an artist. Time constraints mean that unique assets may not be created for every character in the environment, resulting in asset repetition throughout the crowd which must be disguised. Rendering crowds of larger sizes, detail and variety requires increased computation time. Though the creation of a fully populated city for non real time applications such as movies may be done offline, with hours spent rendering each frame, time may be saved by approximating non-real time techniques for real time applications such as games. In a real time application frames must be drawn at a rate of at least 24 per second to maintain visual fluidity.