Making it safer to be in a crowd
The management of events involving large crowds is an art, mainly based on human expertise, which is behind planning the layout of the venue, access arrangements, and the mapping of certain crisis scenarios. Supervision during the event, which aims to make any decisions regarding the actions to be taken to keep the crowd safe and in line with the expected level of service, is also down to human judgement. While tragedies arising from crowd dynamics are rare, they do happen, and when they do, it’s often down to a rapidly changing context that human judgement can struggle to anticipate and keep up with. So how can the latest digital tools be harnessed to make situations safer? Julien Pettre CrowdDNA project coordinator, based at the French National Institute for Research in Digital Science and Technology (Inria), explains: “It seems clear to us that the difficulty of the exercise lies mainly in the manager’s ability to assess the situation correctly. “It is on this last point that technology can intervene and assist a manager in the precise evaluation of a situation and in the detection of clues that may indicate that danger is imminent, especially at a time when human intervention to dispel this risk is still possible.” CrowdDNA, which received support from the EU, focused on situations involving high-density crowds. In this situation, the risks for those in the crowd occur when there is too much density in some places, which prevents individuals from moving freely. “They are trapped, and suffer from the erratic movements of the crowd, which only aggravate the situation. If the density is too high for a long period of time, this can lead to the risk of asphyxiation and death for individuals,” Pettre says.
Revolutionising how we observe dense crowds in real time
Using computer animation, virtual reality, robot navigation and motion planning, CrowdDNA pursued two lines of innovation to improve crowd safety. The team set out to understand the nature and intensity of physical interactions between individuals in crowds and to reproduce these phenomena in digital simulation. And then they aimed to use these simulation tools to create crowd movement analysis technologies that make it possible to detect potentially dangerous situations as early as possible. The project revolutionised the methods used to observe interactions in dense crowds. Whereas video capture and 2D positional reconstruction of individual trajectories had been the norm in the field, CrowdDNA introduced the use of full-body motion capture technologies to study physical interactions between individuals. “This produced unprecedented datasets showing, for example, the propagation of pushes between individuals in a crowd,” Pettre explains. He adds that the project also laid the foundations for capturing data on the ground, through the concept of ‘crowd observatories’, which, he feels, were sorely lacking, even though it is essential that those managing the situation know what is happening in situ. Along with developing innovative ways of observing crowd dynamics as they unfold, the team also focused on modelling and simulation to spot potential dangers. “While simulation models are generally simplistic, reducing individuals to particles moving in one plane, we have laid the foundations for crowd simulation models that capture the complexity of bodily interactions between individuals in contact at limb level. This radically changes the concepts used for crowd simulation and enables the study of dense crowd situations where complex physical exchanges take place,” notes Pettre.
Designing simulations of crowd dynamics to improve analytics
While accidents linked to dense crowds are rare events for which data is generally unavailable, CrowdDNA has laid the foundations for an approach in which simulation is used as the main source for developing techniques for analysing sequences of crowd images. This idea of training situation analysis assistance technologies, solely on the basis of synthetic data, may make it possible in the future to design technologies that can be adapted to very specific cases. As Pettre says: “Let’s not forget that each crowd in different places and circumstances is unique, and that this ability to adapt is invaluable.”
Keywords
CrowdDNA, dense crowds, crowd observatories, computer animation, virtual reality, robot navigation, motion planning, crowd safety
