Characterization of Dense Crowd Using Gibbs Entropy.

Automatic understanding of crowd dynamics through computer vision is a challenging task. There exists a number of works related to crowd behaviors, especially when the gatherings are large. A crowd can be characterized by its speed, randomness, or density. Large gatherings at sociocultural events often cause traffic congestion at cities or even they lead to untoward incidents such as stampede or accidents. However, if the crowd dynamics can be understood or predicted, precautionary measures can be taken by the administrative authority. In this paper, a Gibbs entropy-based crowd characterization has been proposed to estimate crowd dynamics, especially at large gatherings. The average frame energy estimated by the kinetic energy of moving particles has been used to estimate the speed of movement, while the average frame entropy gives information about the randomness in the crowd. The proposed method has been evaluated on two publicly available datasets as well as our dataset videos recorded during sociocultural gatherings. It has been observed that the proposed entropy-energy-based analysis can successfully characterize crowd dynamics and it can be used for flow analysis and density estimation.