Crowd Anomaly Detection in Public Surveillance via Spatio-temporal Descriptors and Zero-Shot Classifier

Anomaly detection has become an imperative defense line in public surveillance systems to avert disasters and to monitor crowd behavior. In this paper, we proposed a robust anomaly detection system using Zero-Shot Learning (ZSL) classifier with hybrid Spatio-temporal descriptors, firstly, we pre-processed extracted video frames then we introduced semantic segmentation for foreground extraction depicting human pixels only after that we extract new robust Spatio-temporal descriptors including Geometric, SIFT, time domain and crowd contour variation descriptors. Next, the Particle Swarm optimization algorithm was used for optimal features selection, and finally, we introduced ZSL for decision-making based on extracted optimal features. We evaluate the performance of our proposed system on two publicly available UMN and MED benchmark datasets and achieved an accuracy rate of 93.5% and 91% respectively. Experimental results reveal that our system achieves superior accuracy as compared to existing well-known state-of-the-art models. The proposed system can be utilized at numerous public places, like train stations, shopping malls, airports, city centers, and university campuses to automatically detect, control, and supervise the crowd.