“Behavior-graph for crowd simulation” by Lerner, Fitusi, Chrysanthou and Cohen-Or

Computer generated crowds can be found in films, commercials, computer games and other virtual environment applications. Usually, crowd simulation techniques, such as [Treuille et al. 2006; Lerner et al. 2007] focus on controlling the direction and speed of the simulated agents, aiming at generating realistic crowds at the trajectory level. In principle, if the simulated paths reflect paths taken by real people, then the agents would appear to behave correctly. However, the definition of behavior does not end at the trajectory level. It can be defined as the aggregate actions of a person to internal or external stimuli. Talking to nearby people, or glancing at something of interest are common actions performed by a person in a crowd. Actions such as these, which are often missing from simulated crowds, are vital for generating seemingly natural behaviors. Ideally, actions should be generated by the crowd simulation algorithm itself. However, in order to enrich existing simulations, one can examine its output, deduce the stimuli of each simulated agent in the scene and assign plausible actions that are compatible with their trajectory. In this sketch we describe a technique for adding behavior information to simulated crowds as can be seen in the accompanying video.

1. Lerner, A., Chrysanthou, Y., and Lischinski, D. 2007. Crowds by example. to appear in EG2007.
2. Treuille, A., Cooper, S., and Popovic, Z. 2006. Continuum crowds. ACM Trans. Graph. 25, 3, 1160–1168.