Real-Time Anomaly Detection in CCTV Surveillance using Deep Learning

Due to the increase in crime in public spaces, there is a growing need for automated detection systems that are capable of detecting abnormal behaviour of humans by analysing footage from surveillance cameras. Manual observation is time consuming, expensive and often full of human error. This paper describes a Hybrid Deep Learning Framework comprised of implementing YOLOv8 object detection along with implementing a CNN for identifying suspicious behaviour, which utilises the UCF-Crime dataset to assess behaviours in nine different activity categories. YOLOv8 provides real-time object detection, while the CNN classifies the detected behaviours as either normal or abnormal based upon extracted spatial features from video frames. The Hybrid Deep Learning Framework produced an overall accuracy of 98.76% when utilising an 80/20 train/test split, with high precision, recall and F1- scores indicating that real-time surveillance can be performed effectively by the Hybrid Deep Learning Framework and that this level of real-time surveillance supports improved public safety.