SPECIFICATION OF THE SOCIAL FORCE PEDESTRIAN MODEL BY EVOLUTIONARY ADJUSTMENT TO VIDEO TRACKING DATA

JOHANSSON, ANDERS; HELBING, DIRK; SHUKLA, PRADYUMN K. · 2007 · OpenAlex-citations

DOI: 10.1142/s0219525907001355

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of calibrating microscopic pedestrian simulation models, specifically the social force model, to ensure realistic and reliable results for large-scale applications. While previous methods often adjusted parameters to fit fundamental flow-density diagrams, this approach is insufficient for reproducing complex dynamical phenomena such as lane formation, stripe patterns, and bottleneck oscillations. The authors propose a novel calibration method that uses evolutionary optimization to fit model parameters directly to empirical trajectory data extracted from video recordings of pedestrian crowds. The methodology involves recording pedestrian motion in natural environments, including escalator entrances, shopping malls, and pedestrian crossings in Budapest and Stuttgart. Automated tracking software identified pedestrian heads and projected their trajectories onto a two-dimensional plane. To calibrate the model, the authors employed a hybrid simulation where virtual pedestrians moved according to the social force model while surrounding pedestrians followed the empirically tracked trajectories. An evolutionary algorithm optimized the interaction parameters—specifically the interaction strength ($A$), interaction range ($B$), and angular dependence ($\lambda$)—by minimizing the relative distance error between simulated and tracked positions over short time intervals. The study compared three interaction force specifications: a circular model, and two elliptical models that account for velocity-dependent interactions. The results demonstrated that the elliptical specification II, which symmetrically treats both interacting pedestrians and accounts for their velocities, yielded the best fitness value of -0.39, significantly outperforming the circular model (-0.60) and the extrapolation baseline (-0.66). The optimal parameters for this model were determined to be $A = 0.04$, $B = 3.22$, and $\lambda = 0.06$. The calibrated model successfully reproduced the empirical Weidmann fundamental diagram for pedestrian flow. Furthermore, data-driven analysis of interaction forces revealed that pedestrians are primarily sensitive to stimuli within a 180-degree angle in front of them, and the distance dependence of interaction forces follows an exponential decay. The significance of this work lies in providing a robust, data-driven method for specifying microscopic pedestrian models, moving beyond qualitative adjustments. The calibrated model was validated through large-scale simulations, including evacuation scenarios for ships and exhibition halls, and massive crowd dynamics such as the Hajj pilgrimage at the Jamarat Bridge. The authors also introduced "UNIVERSE," a scalable simulator capable of handling over 100,000 pedestrians in real-time using parallel computing. This framework enables accurate assessment of pedestrian facility designs and safety measures in complex urban and emergency environments.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• pedestrian behavior perception