Gast ::
| Hauptseite > Publikationsdatenbank > Combining socio-psychological pedestrian dynamics with AI and simulation tools |
| Hauptseite > Publikationsdatenbank > Combining socio-psychological pedestrian dynamics with AI and simulation tools |
| Poster (After Call) | FZJ-2025-05612 |
; ; ; ;
2025
This record in other databases:
Please use a persistent id in citations: doi:10.34734/FZJ-2025-05612
Abstract: In the Pedestrian Dynamics – Social Psychology division at the IAS-7 we investigate socio-psychological processes that influence pedestrian dynamics, with a focus on how crowds and pedestrian infrastructures are perceived and experienced. In our research we combine quantitative and qualitative methods to explore a range of social-psychological phenomena in crowds, including anonymity, social norms, and emotional and motivational dynamics. Our data collection methods include questionnaires, psycho-physiological measures (e.g., heart rate, skin conductance, gaze), and observational data obtained through experiments, field studies, and interviews. We maintain close interdisciplinary collaborations with colleagues from physics, mathematics, and computer science, and are actively involved in the development of pedestrian models and simulations, aiming to incorporate socio-psychological factors into collective dynamics.To give an example of the interdisciplinary focus of our work at the IAS retreat in May, we have selected a specific research topic. This study combines psychological and computational approaches, offering a unique perspective on pushing behaviour in crowds and highlighting the strength of interdisciplinary collaboration. We investigated the conditions under which individuals in a crowd begin to intentionally push on two different levels. On a social level, the question arises as to whether such behaviour spreads through social influence. On an individual level, the role of motivation is to be studied. To address these questions, we developed a manual rating system to code the intensity of forward motion on a 4-point ordinal scale. Applying this method to a set of laboratory experiment datasets allowed us to examine the relationship between spatio-temporal factors (e.g., proximity to the goal) and pushing intensity, as well as the local spread of pushing through neighborhood analysis. Building on these findings, we explored the feasibility of automatically detecting pushing behavior using AI. Several AI systems were trained on the rated video data to identify pushing behavior: First in recorded footage with rough location estimations, then in live camera streams with the same approach, and finally at the level of individual actors, pinpointing those engaged in pushing. Ultimately, this collaboration demonstrates a flow from social psychological coding to the development of a practical tool for crowd management. This tool can be used in real-time interventions, allowing managers to respond (e.g., enlarging areas where pushing occurs) to maintain safety and flow.
|
The record appears in these collections: |
PDF