Pedestrian trajectories are used in urban and behavioral research to learn about human behavior in public space and the impact of spatial features on pedestrian flows. Currently, these trajectories are collected manually, with self-tracking devices, or with video cameras. Even when trajectories are obtained using computational techniques, such as using computer vision to trace them in space, these datasets are not made available for reproducibility or comparative studies between different locations. To close this gap, this dataset makes available the data of pedestrian trajectories collected in 39 European squares. The collection process is based on collecting footage from publicly available webcams. Trajectories where then extracted using object detection (YOLOv10), tracking (Bytetrack), and georeferencing (QGis and Python-opencv). Data cleaning and validation steps are described in the README file. The dataset ultimately includes 348,200 pedestrian trajectories extracted from 193 hours of video footage, collected at different times of the day, during working days and weekends, and during the Spring and Summer season.
Wolff et al. (Thu,) studied this question.