The internet of things for crowd panic detection

Abstract

Crowd behavior detection is important for the smart cities applications such as people gathering for different events. However, it is a challenging problem due to the internal states of the crowd itself and the surrounding environment. This paper proposes a novel crowd behavior detection framework based on a number of parameters. We first exploit a computer vision approach based on scale invariant feature transform (SIFT) to classify the crowd behavior either into panic or normalness. We then consider a number of other parameters from the surroundings namely crowd coherency, social interaction, motion information, randomness in crowd speed, internal chaos level, crowd condition, crowd temporal history, and crowd vibration status along with time stamp. Subsequently, these parameters are fed to deep learning model during training stage and the behavior of the crowd is detected during the testing stage. The experimental results show that proposed method renders significant performance in terms of crowd behavior detection.