In-depth: Interdisciplinary investigations of pushing in dense crowds

In many dangerous crowd situations, the combination of high density (more than 8 persons per m²) and movement within the crowd results in strong pressure on the human body and increases the likelihood of life-threatening falls. In our research, we have reconstructed this dynamic for the Love Parade disaster in 2010 based on a large sample of witness reports:

Inside a life-threatening crowd: Analysis of the Love Parade disaster

This study has highlighted the necessity for in-depth research into the role of intentional pushing, impulse propagation and pressure in very dense crowds. We have launched a comprehensive series of scientific studies using methods from psychology, physics and computer science to answer 3 key research questions (published in 9 peer-reviewed articles):

1. How is impulse propagated in a crowd? What collective dynamics result from pushing in the crowd?

A series of controlled laboratory experiments investigated the impact of an external impulse on different formation of small or large groups (standing in a row or a huddle, with varying densities). Measurements of head trajectories and the use of 3D motion capturing suits and pressure pads enabled us to study the role of the human body within the propagation of impulse.

Journal articles:
Forward propagation of a push through a row of people
Propagation of controlled frontwardimpulses through standing crowds
Temporal segmentation of motion propagation

2. Under which circumstances do people in a crowd start to push intentionally? Does pushing behavior propagate so that more and more people join in?

In order to address these psychological questions within the context of motivation and social influence, a manual rating system was developed to code the intensity of forward motion, including pushing, on a 4-point ordinal scale. Applying this rating method to a series of laboratory experiments allowed us to calculate the relationship between the spatial and temporal dynamics of the crowd (i.e. how close someone is to the goal) and the intensity of pushing. Furthermore, a neighborhood analysis looked at the social influence of pushing and the spread of this behavior.

Journal articles:
Psychological pushing propagation
Pushing and Non-pushing Forward Motion in Crowds
Dynamic Relationship between Pushing Behavior and Crowd Dynamics

3. Is automatic detection of pushing in crowds via AI possible?

On the basis of the rated experiment material (intensity of forward motion, see 2.) several AIs were systematically trained to detect pushing within video material. In phase one, the system was trained to automatically identify regions containing individuals engaged in pushing behavior from video recordings of crowds. In the second phase, the same detection method was used, but the AI was trained to operate on live camera streams. In the final phase, the system was developed to focus on individuals directly, detecting persons who engage in pushing behavior individually.

Journal articles:
A Voronoi-based convolutional neural network for pushing
A Cloud-Based Deep Learning Framework for Detection of Pushing
A Fast Hybrid Deep Neural Network Model for pushing detection