Learning Behavior Patterns from Video: A Data-driven Framework for Agent-based Crowd Modeling

Abstract

This paper proposes a generic data-driven crowd modeling framework to generate crowd behaviors that can match the video data. The proposed framework uses a dual-layer mechanism to model the crowd behaviors. The bottom layer models the microscopic collision avoidance behaviors, while the top layer models the macroscopic crowd behaviors such as the goal selection patterns and the path navigation patterns. Based on the dual-layer mechanism, an automatic learning method is proposed to learn the model components from video data. To validate its effectiveness, the proposed framework is applied to generate the crowd behaviors in New York Grand Central Terminal. The simulation results demonstrate that the proposed method is able to construct effective model that can generate the desired emergent crowd behaviors and can offer promising prediction performance.