Dufour Oscar, Dang Huu-Tu, Cordes Jakob, Korbmacher Raphael, Chraibi Mohcine, Gaudou Benoit, Nicolas Alexandre, Tordeux Antoine
Université Claude Bernard Lyon 1, CNRS, Institut Lumière Matière, UMR5306, F-69100, Villeurbanne, France.
UMR 5505 IRIT, Université Toulouse Capitole, Toulouse, France.
Sci Data. 2025 Apr 30;12(1):718. doi: 10.1038/s41597-025-04732-3.
The dynamics of dense crowds have received considerable attention from researchers seeking fundamental understanding or aiming to develop data-driven algorithms to predict pedestrian trajectories. However, current research mainly relies on data collected in controlled settings. We present one of the first comprehensive field datasets describing dense pedestrian dynamics at different scales, from contextualized macroscopic crowd flows over hundreds of meters to microscopic trajectories (around 7000 individual trajectories). In addition, a sample of GPS traces, some statistics of contacts and pushes, and a list of non-standard crowd phenomena observed in the video recordings are provided. The data were collected during the 2022 Festival of Lights in Lyon, France, as part of the French-German MADRAS project and cover densities up to 4 pedestrians per square meter. We suggest using this extensive dataset, acquired in complex real-world settings, to benchmark models of pedestrian dynamics.
密集人群的动态受到了研究人员的广泛关注，他们旨在深入理解这一现象，或者开发数据驱动的算法来预测行人轨迹。然而，目前的研究主要依赖于在受控环境中收集的数据。我们展示了首批全面的实地数据集之一，该数据集描述了不同尺度下的密集行人动态，从数百米范围内的情境化宏观人群流动到微观轨迹（约7000条个体轨迹）。此外，还提供了一组GPS轨迹样本、一些接触和推搡的统计数据，以及视频记录中观察到的一系列非标准人群现象。这些数据是在法国里昂2022年灯光节期间收集的，作为法德MADRAS项目的一部分，涵盖了每平方米多达4名行人的密度。我们建议使用这个在复杂现实世界环境中获取的广泛数据集，来对行人动态模型进行基准测试。
