About crowd dynamics

General public poster version  
Technical poster 

Pedestrians walking in crowds show complex dynamics and stunning collective motion.
The investigation of the underlining behaviors from single individuals to large-scale interactions is the topic of our research.

The typical questions we challenge ourselves with are

  • Can we write quantitative models for the behavior of pedestrians walking in crowds?
  • Are pedestrian crowds moving like fluid flows?

Answers to these questions has primary impact in engineering, as they enable accurate evacuation simulations, crowd management and interactive built environments.

Figure 1: The main walkway of Eindhoven Train Station. Four overhead Kinect sensors record continuously the traffic flow.

High statistics measurements

Walking individuals have an intrinsically random behavior. A quantitative prediction of walking trajectories is thus impossible. Nevertheless, we expect walking pedestrians to exhibit universal features at the statistical level. In other words, we expect the distributions of position, velocity or acceleration to describe the dynamics.
Comparison with measurements is paramount to shed light on this aspect. Hence, we perform extensive and highly accurate experimental measurements of pedestrian trajectories. To explore the statistical portrait including the rare events we perform 24/7 automatic data acquisition on a yearly basis. We established different measurement locations within TU/e [1, 2] and at Eindhoven Train Station (cf. Figure 1 and 2).

Figure 2: A depth map reconstructed from the recordings of four Kinect sensors. Darker pixels are closer to the plane of the cameras. Pedestrian trajectories and bodies orientation are estimated via an ad hoc method.

3D-range sensors for accurate tracking

To measure pedestrian trajectories accurately we employ an ad hoc technique utilizing multiple overhead Microsoft Kinect 3D-range sensors [1,2]. 3D-range sensors deliver depth maps that associate to each point its distance with the camera plane (cf. Figure 2). We perform pedestrian segmentation via clusterization algorithms applied on the 3D data. Thus, analyzing each cluster pedestrian we identify the head, which we track as a particle in a flow.

Quantitative modeling

We derive Langevin-like stochastic particle models to reproduce the dynamics of pedestrians. We deduce potentials in space and velocity to mimic and explain the observed non-Gaussian fluctuations up to rare events [1].



Alessandro Corbetta; Chung-min Lee; Roberto Benzi; Adrian Muntean; Federico Toschi

Fluctuations around mean walking behaviours in diluted pedestrian flows Journal Article

Physical Review E, 95 , pp. 032316, 2017.

Abstract | Links | BibTeX


Alessandro Corbetta; Luca Bruno; Adrian Muntean; Federico Toschi

High statistics measurements of pedestrian dynamics Journal Article

Transportation Research Procedia, 2 , pp. 96-104, 2014.

Abstract | Links | BibTeX