An Application of Graphics Processing Units to Geosimulation of Collective Crowd Behaviour

The goal of the paper is to assess the ways for computational performance and efficiency improvement of collective crowd behaviour simulation by using parallel computing methods implemented on graphics processing unit (GPU). To perform an experimental evaluation of benefits of parallel computing, a new GPU-based simulator prototype is proposed and the runtime performance is analysed. Based on practical examples of pedestrian dynamics geosimulation, the obtained performance measurements are compared to several other available multiagent simulation tools to determine the efficiency of the proposed simulator, as well as to provide generic guidelines for the efficiency improvements of the parallel simulation of collective crowd behaviour.