AN INTELLIGENT AI-DRIVEN SURVEILLANCE FRAMEWORK FOR REAL-TIME ANOMALY DETECTION

Abstract

The project introduces a CCTV surveillance system driven by AI that can monitor intelligently and in real time with little assistance from humans. The continual manual observation that is a major component of traditional surveillance is prone to tiredness and delayed response. A clever and proactive security solution appropriate for continuous round-the-clock monitoring is made possible by the suggested system, which combines computer vision and deep learning technologies to automatically recognize, track, and evaluate activities in video footage. The system makes sure that people and objects are correctly detected and reliably tracked across video frames by using YOLOv8 for real-time object detection and DeepSORT for multi-object tracking. Because of this combination, the system can keep an eye on movement patterns and preserve individual identities even in environments that are busy or dynamic. CCTV footage, uploaded movies, or live streams may all be used as video input. Python and OpenCV are used for processing, and the interface is straightforward and Streamlit-based. The project uses a rule-based anomaly detection layer built on top of detection and tracking outputs in place of a complex training-based anomaly model. Based on motion patterns and object counts, it detects suspicious circumstances such crowd gathering, unusually rapid movement, and intrusion. Operators may quickly and easily assess events thanks to the visual cues offered by the red bounding boxes for suspicious activity and the green bounding boxes for normal activity. Using the Brevo SMTP API, the system instantly sends out email alerts when an anomaly is found, guaranteeing prompt reporting to the relevant authorities. The framework runs on common systems without the need for specialist hardware, is entirely implemented in Python, and is portable via a virtual environment. This study shows how an intelligent, scalable, and effective surveillance solution for contemporary security applications may be produced by integrating AI, object identification, tracking, and automated warnings.