Scalable pedestrian simulation for virtual cities

摘要: Most of the common approaches for pedestrian simulation, used in Graphics/VR community, are bottom-up. The avatars individually simulated space and overall behavior emerges from their interactions. This can lead to interesting results but it does not scale be applied populating a whole city. In this paper we present novel method that scene almost any size. We use top-down approach where movement pedestrians is computed at higher level, taking global view model, allowing flux densities maintained very little cost city level. information stochastically guiding more detailed realistic low level simulation when user zooms specific region, thus maintaining consistency.At heart system an iterative models flow as random walk. People moved around graph nodes until model reaches steady state which provides feedback avatar navigation run time. Negative Binomial distribution function number people leaving each node while selected direction based on popularity through preference-factor. preference-factor parameters including visibility node, events place so on.An important feature low-level dynamics interactively specify intuitive variables predictably modify collective region; density, selectively modified.