Direction, velocity, merging probabilities and shape descriptors for crowd behavior analysis

Abstract

Descriptors are important for quantifying crowd behavior. The existing descriptors generally provide information about crowd density, i.e., number of people/objects present in a defined spatial area. However, other properties of crowd like speed, direction, shape, and merging probabilities (of different crowds at group level) are also important for crowd analysis. In this paper, crowd descriptors (by mitigating the effects of outliers) are introduced which can be used for crowds having various densities. The simulations on various datasets show the applicability of the proposed descriptors.