Overview: The video surveillance system market has become popular and used in public places such as shopping malls, hospitals, banks, streets, educational institutions, municipal governments, smart cities, etc. to improve the security of public life and assets . Timely and accurate detection of video anomalies is almost always the primary goal of any security application. A video anomaly, such as anomalous activity or anomalous entity, is defined as an anomaly or irregular pattern present in the video that does not correspond to the normal trained pattern. Unusual activities such as brawls, riots, traffic violations, mass panics, and unusual entities such as weapons and misplaced luggage in sensitive locations should be automatically detected in time. However, it is difficult to detect video anomalies due to the ambiguity of anomalies, various environmental conditions, the complex nature of human behavior, and the lack of suitable datasets. This article focuses on the evolution of anomaly detection, followed by an overview of the various methods developed to detect anomalies in intelligent video surveillance. Additionally, it leverages the last decade of anomaly detection research. The following is a systematic taxonomy of anomaly detection methods. Since the concept of anomalies is contextual, anomaly detection identifies different objects of interest and published datasets. Since anomaly detection is a time-sensitive application of computer vision, we explore anomaly detection using edge devices and approaches designed explicitly for it. We also explore the confluence of edge computing and anomaly detection for real-time, intelligent surveillance applications. It also discusses the challenges and opportunities of anomaly detection.
Keywords: Detection, Anomaly, Normal-abnormal, Neural-Network
|
DOI:
10.17148/IJARCCE.2023.12111
