SocioCrowd: a social-network-based framework for crowd simulation

The goal of crowd simulation is to produce potential collective behaviors by simulating the movement process of a number of characters or agents. Some famous models are proposed to simulate crowd, including social force (e.g. [Helbing 2000]), cellular automata (e.g. [Chenny 2004]), and rule-based models (e.g. [Reynolds 1987]). Others use physiological (e.g. locomotion, energy level) and psychological (e.g. impatience, personality attributes) traits of agents to trigger heterogeneous behaviors [Pelechano 2007]. However, existing approaches do not consider the real-world social interactions among agents, and thus are unable to produce social-dependent scenarios. In this work, we propose to leverage the underlying social network, which captures social relationships among agents, for crowd simulation. A novel social-network-based framework, SocioCrowd, is developed (figure 1(a)) shows the virtual world). Based on SocioCrowd, we simulate three social-based scenarios, including community-guided flocking, following leading persons, and spatio-social information spreading. They display certain real-world social behaviors which are hardly modeled by existing methods. To lift the performance, our SocioCrowd is implemented by pure Java with GPU programming in ways of GSGL and JCUDA.