Anomaly Detection in Video Surveillance using Deep Learning Techniques: A Review

Video surveillance plays a vital role in ensuring public security with the application of computer vision technologies to analyze and recognize long video streams. Anomaly detection in video surveillance is more challenging due to the complex motion patterns, ambiguous nature of the anomaly, different environmental conditions, and the lack of proper datasets. Recently, more anomaly detection systems have been developed to automatically detect the anomalous events in the video streams. Nevertheless, the state-of-the-art reviews do not offer a thorough analysis that covers all aspects including methods, modeling, performance evaluation techniques, and trending research issues related to video anomaly detection. From this perspective, the main contribution of the proposed review is to provide a comprehensive study of various learning-based techniques in the field of anomaly detection. In addition, the review describes the benchmark datasets that are used in the detection process and discusses the advances and challenges in the existing methods precisely. More specifically, this review analyzes 25 recent literature works with diverse techniques and approaches that are utilized in the field of anomaly detection, which are briefly explained in the review article. Additionally, the common limitations of surveillance systems are summarized and pose some possible directions for future studies. Based on this information, the study explores more ideas about anomaly detection and offers promising directions for future research to advance the field of video surveillance.