Real Time Visualization of Crowd Dynamics Scenarios

This paper presents an approach to real-time simulation of crowd dynamics on GPU enabled computing architectures. We discuss challenges with parallelization of agent-based models, implementing parallel simulation algorithms, visualization and interaction with the simulated scene and, most importantly, ensuring communication and synchronization between all these processes. Our main objective is to provide interactive simulation of realistic models such as pedestrian dynamics, in which large crowds move and interact among themselves and with the environment. Simulation parameters like scene complexity, scene composition, as well as the number of agents are varied in order to simulate different scenarios and to assess the impact on performance.