Abstract

AI in video surveillance has been a game changer, elevating systems into a new industry, widely known as smart monitoring systems. This review summarizes recent developments in the domain of AI-based video surveillance systems, emphasizing the convergence of object detection, face detection, and face recognition. In particular, it expounds on real-time human detection, explains deep learning techniques including YOLO, facial localization by MTCNN and RetinaFace, and identity verification by ArcFace and FaceNet. This paper discusses the relative merits and drawbacks of these approaches, how they have been employed in practice, and the ethical and societal implications they raise. The goal of this review is to summarize and analyze the state of research, describe a selection of performance benchmarks, spot notable gaps in the literature, and recommend paths for future development in automated suspect identification.