Computer Vision Techniques for Automated Surveillance Systems
Modern security systems have raised a new system of a need to integrate automated surveillance systems as part of their security infrastructure because of the sudden increase in urbanization, civil safety issues and the necessity to have a smart monitoring system. The conventional surveillance systems are very dependent on human operator hence constraints include fatigue, delay in response and subjectivity. Computer vision as a branch of artificial intelligence will allow machines to read and understand visual information automatically, and thus change the traditional surveillance into its intelligent and active form. This paper gives an extensive research of the computer vision approaches in automated surveillance systems. It also explores how the classical approaches to image processing have been transformed to deep learning based methods such as their application in object detection and tracking, activity recognition, anomaly detection and facial recognition. System architectures, data acquisition pipelines, feature extraction methods, model training strategies and performance evaluation metrics are also discussed in the paper. Moreover, the issues like occlusion, change of illumination, scalability, privacy, real-time processing are examined. The effectiveness of the modern computer vision methods is discussed with references to the experimental results of the representative surveillance scenarios. Lastly, the paper provides the future research direction, such as edge-AI surveillance, multimodal fusion, and explainable computer vision, which is important to the next-generation intelligent surveillance systems.