IAS-Seminar "High statistics measurements of pedestrian dynamics and modelling"

Referent:

Dr. Alessandro Corbetta, Department of Applied Physics, Eindhoven University of Technology

Abstract:

The dynamics of pedestrian crowds is a relevant topic for the design and safety of civil infrastructures and furthermore a fascinating subject deeply connected with many scientific disciplines, including statistical physics and fluid dynamics. Walking pedestrians exhibit a large variety of dynamical behaviors, possibly influenced by the surrounding crowd and by the environment. Even in very simple geometries, individuals always display, in addition to average behaviors, small fluctuations and, rarely, large ’anomalous’ deviations. Observations of crowds with very high statistics are thus expected to be able to display the statistical signature of both frequent and rare fluctuations. Quantifying and modeling such signature is hence a paramount step to increase our understanding of the physics of crowds and our quantitative simulation capabilities.

To investigate the statistics of fluctuations and of rare events, in 2013 we started different year-long experimental campaigns in real-life settings. We collected on a 24/7 basis the trajectories of pedestrians walking in different locations, namely a corridor within Eindhoven University of Technology, the main walkway of Eindhoven train station and during the Eindhoven 2016 Glow festival. The heart of this data collection are grids of overhead Microsoft Kinect 3D range sensors and ad hoc detection and tracking algorithms. These enabled reliable and highly accurate measurements, unprecedented in scale, both offline and, more recently, in real-time.

First, we will discuss the measurement setups and techniques along with the challenging problem of identifying similar flow conditions occurring randomly within our measurements. This is a necessary step to ensure coherent statistical analysis. Then we focus on diluted flow conditions: considering pedestrians walking undisturbed, we discuss a Langevin-like stochastic model recently proposed by us reproducing quantitatively measured statistics of position and velocity, as well as rare velocity inversion events. Thus, we complement the model with mutual avoidance terms comparing their statistical properties with the experimental data.

Alle Interessierten sind zu diesem Vortrag herzlich eingeladen.

Kontakt: Dr. Maik Boltes, JSC