Improving crowd behaviour for emergency simulation using game-captured data.

Abstract

Crowd simulation has become an essential asset in many different industries, ranging from games and movies to engineering and construction; while the requirements might vary significantly between these domains, the underlying theory is shared among them. Emergency simulation in particular is a field that has made great use of crowd simulation to validate and improve building design and layouts, allowing architects to determine how a building would cope in case of an evacuation and determine the estimated egress time and survival rate. Existing algorithms and techniques are usually organised in two layers: a high level component determines the actions and goals of each agent that is participating in the simulation while the low level component is responsible for the movement of the agent. Steering behaviours [Reynolds 1987] are usually employed for movement, but they don’t take into account more nuanced aspects of agents behaviour like their internal state, priorities and system of beliefs [Pelechano et al. 2007]. This is particularly important for emergency simulations as individuals react in different ways under stressful circumstances and factors like fatigue and caring for others must be taken into account to provide an accurate simulation.