People Flow Prediction by Multi-Agent Simulator

In places where many people gather, such as largescale event venues, it is important to prevent crowd accidents from occurring. To that end, we must predict the flows of people and develop remedies before congestion creates a problem. Predicting the movement of a crowd is possible by using a multi agent simulator, and highly accurate prediction can be achieved by reusing past event information to accurately estimate the simulation parameters. However, no such information is available for newly constructed event venues. Therefore, we propose here a method that improves estimation accuracy by utilizing the data measured on the day of the event. We introduce a people-flow prediction system that incorporates the proposed method. In this paper, we introduce results of an experiment on the developed system that used people flow data measured at an actual concert event.