Enhancing anomaly detection in video surveillance with spatio-temporal enhanced deep associative memory networks
In the period of extensive video data generation from various surveillance sources, ensuring public safety and security is vital. Detecting unusual crowd behavior is essential, especially in scenarios with large gatherings. However, despite widespread video surveillance, incidents like vehicular accidents, stampedes, and burglaries still occur due to the limitations of traditional surveillance systems. In anomaly detection, finding the critical deviated pattern is the main and critical task. In the context of intelligent video surveillance, automated detection of abnormal behavior is achieved through computer vision analysis, eliminating the need for constant human monitoring. This paper proposes the spatio-temporal enhanced deep associative memory networks (STEAD)-network, a novel approach for anomaly detection in video sequences. The STEAD-network combines various techniques, including spatio-temporal enhancement, associative memory modules, and pattern recognition, to effectively capture and recognize abnormal events. Three benchmark datasets, UCSD Ped2, CUHK Avenue, and ShanghaiTech are used to evaluate this proposed dataset and to compare with existing state-of-the-art techniques. The results demonstrate that the STEAD-network consistently outperforms other methods, achieving significant improvements in anomaly detection accuracy across all datasets. The development of intelligent video surveillance systems is aided by this research by enhancing their ability to autonomously and accurately detect abnormal behavior in real-world scenarios.