Understanding Collective Crowd Behaviors:

Learning Mixture Model of Dynamic Pedestrian-Agents

Bolei Zhou1, Xiaogang Wang2,3, and Xiaoou Tang1,3

1Department of Informaiton Engineering, 2Department of Electronic Engineering, The Chinese University of Hong Kong

3Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences


[Paper] [Simulation Video] [Dataset] [Presentation]

1. Motivation

Collective crowd behaviors widely exist in nature. Here are some examples of collective crowd behaviors:

To understand these collective crowd behaviors, we need to figure out the unique features of collective crowd behaviors. From the study of social science, there are three critical features of collective crowd behaviors:

In our project, we would like to understand the collective crowd behaviors in New York Grand Central Station, and build computational model to learn these collective motion patterns from highly fragmented trajectories.

However, there are a lot of challenges such as 1) different collective patterns mixed together 2) high detection and tracking errors due to crowdedness.

2. Framework of Dynamic Pedestrian-Agents

To learn the collective behavior patterns from highly fragmented trajectories, we propose the framework of Dynamic Pedestrian-Agents.

Three contributions of this framework are:

  • Agent-based modeling of crowd behavior: Every pedestrian is driven by on type of agents, and the whole crowd is modeled as a mixture of pedestrian-agents. Here are 3 examplar pedestrian-agents:
  • Three factors to analyze pedestrian-agents: 1) Belief of pedestrian (starting point and destination) 2) Collective Dynamics(pedestrian movement patterns) 3) Timing of emerging (the frequency of one type of pedestrians entering the scene)
  • Handling highly fragmented trajectories

Mathematical modeling of each components and the graphical model of Dynamic Pedestrian-Agents are


3. Experimental Results

The framework of Dynamic Pedestrian-Agents has a lot of applications:

Simulation Video of Dynamic Pedestrian-Agents


4. Reference

B. Zhou, X. Wang and X. Tang. "Understanding Collective Crowd Behaviors: Learning a Mixture Model of Dynamic Pedestrian-Agents." in Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2012 


Last update: July 4, 2012