When people gather in large groups like those found at Black Friday sales events, pilgrimages, concerts, and parades, crowd density often becomes exceptionally high. As a consequence, these events can produce tragic crowd disasters such as stampedes and mass crushes. Crowd turbulence, in particular, is due to the dominance of physical interactions between contacting bodies occurring at high density. It is described as the emergence of density waves that suddenly displace people over long distances, possibly crushing them into barriers, causing falling, extreme compression, and death from asphyxia. While human collective motion has been studied with active particle simulations, the underlying mechanisms for emergent behaviour are less well understood, and the frequency of mass incidents calls for immediate research in this area.
In a recent work [Bottinelli et al. PRL 2016], we showed that the techniques developed in the context of jammed granular materials and condensed matter physics can be adapted to predict the emergence of collective motions in simulations of high-density crowds. We modelled individuals as active soft particles gathering at a point of common interest, a situation often recurring during mass gatherings. By applying mode analysis to individuals’ trajectories, we found evidence for Goldstone modes, soft spots, and stochastic resonance, which may be the preferential mechanisms for dangerous emergent collective motions in crowds.
Following up these promising results, our current project is strongly focused on the analysis of real crowds video data. Thanks to an ongoing collaboration with the Event planning department of the City of Stockholm, we will be able to film the crowds gathering at the concerts within the Kulturfestival in Stockholm (15-20 August 2017). This will constitute a remarkable dataset, with hours of footages captured by high-definition cameras. By mean of tracking algorithms, we plan to extract the trajectories of individuals and the time evolution of the crowds’ displacement vector field. On the one hand, we will perform mode analysis on both video data and numerical simulations of our crowd model, with particles placed at the same positions where people appear in videos. By comparing the emergent mechanisms extracted from real crowds with the ones from our numerical model, we intend to test the predictive power of mode analysis, which is a decisive step in the validation of mode analysis techniques.
On the other hand, we are developing new analysis tools inspired to condensed matter physics (as correlation analysis) to characterise the collective behaviour of crowds through time series. By matching these new measures with the results of mode analysis on video data, we aim at identifying the signatures of dangerous collective motions. We are currently developing this approach on YouTube videos capturing high-density crowds at concerts.
The results of this project will significantly impact on the understanding of crowd behaviour, with applications in the development of new technologies for real-time risk assessment. On the theoretical standpoint, the project will bear advances in the field of soft active materials, developing new analysis tools merging traditionally unrelated fields of physics.