Insights of Deep Learning-Based Video Anomaly Detection Approaches

Abstract

Deep learning is a powerful computing strategy that has changed the landscape of computer vision. It has been used to tackle complicated cognitive tasks such as detecting abnormalities in videos. Anomalies in the video are events or objects in the footage that don’t fit the typical, learned patterns. Using deep learning, it is possible to automatically and in real-time identify unusual actions and objects like fights, riots, traffic rule violations, abrupt rushes, and the presence of weapons in restricted areas or abandoned luggage. Despite the challenges posed by video anomaly detection, this review offers a comprehensive assessment of published deep learning algorithms for the task. Future research can build on this work by understanding the existing methods to create more effective solutions. First, the challenges of video anomaly identification are discussed as the benefits of deep learning in anomaly detection. Furthermore, several types of abnormalities were explored, followed by diverse methodologies for anomaly identification. Furthermore, significant aspects of anomaly detection using deep learning, including learning approaches, were presented. Finally, numerous datasets used in anomaly detection were examined, followed by a discussion of deep learning-based algorithms for spotting video anomalies.