The role of fluctuations and interactions in pedestrian dynamics

Understanding quantitatively the statistical behaviour of pedestrians walking in crowds is a major scientific challenge of paramount societal relevance. Walking humans exhibit a rich (stochastic) dynamics whose small and large deviations are driven, among others, by own will as well as by environmental conditions. Via 24/7 automatic pedestrian tracking from multiple overhead Microsoft Kinect depth sensors, we collected large ensembles of pedestrian trajectories (in the order of tens of millions) in different real-life scenarios. These scenarios include both narrow corridors and large urban hallways, enabling us to cover and compare a wide spectrum of typical pedestrian dynamics. We investigate the pedestrian motion measuring the PDFs, e.g. those of position, velocity and acceleration, and at unprecedentedly high statistical resolution. We consider the dependence of PDFs on flow conditions, focusing on diluted dynamics and pair-wise interactions ("collisions") for mutual avoidance. By means of Langevin-like models we provide models for the measured data, inclusive typical fluctuations and rare events.