The dynamics of pedestrian crowds is a challenging subject that shares deep connections with statistical physics and fluid dynamics. A key feature of pedestrian dynamics is the intrinsic variability that we can observe already at the single individual level. In this work we aim at providing a quantitative characterisation of this statistical variability by studying individual fluctuations.
Our study is based on the experimental observation of low density pedestrian flows in a corridor of a building at Eindhoven University. Thanks to a year-long 24/7 recording campaign based on Microsoft Kinect 3D-range sensor, we collected an unprecedented statistical database of highly-resolved pedestrian trajectories. In our setting the phenomenology is rather simple: pedestrians walk between opposite ends of the corridor and, rarely, they perform U-turn inversions. We show that we can quantitatively reproduce the observed dynamics and in particular the statistics of ordinary pedestrian fluctuations as well as the rare U-turn events. To this purpose we developed a model based on a generalised Langevin equation that shares similarities with analytic models proposed for active Brownian particles.