Accident and Fight Detection Using Artificial Intelligence

The rapid expansion of CCTV surveillance systems has led to an overwhelming volume of video data, making manual analysis of critical incidents such as road accidents and violent activities inefficient and time-consuming. This paper presents an intelligent video analysis system that automatically detects accident and fight-related anomalies from CCTV footage to support post-incident investigation. The system utilizes a custom-trained model based on YOLOv8 to identify anomalous events within video streams. Upon detecting an accident, vehicle number plates are extracted using optical character recognition (OCR) and linked with temporal information. A multi-camera analysis approach is then employed to trace the movement of the involved vehicles across different camera feeds, enabling reconstruction of their route using timestamp correlations and camera metadata. Experimental evaluation demonstrates that the proposed system effectively detects anomalous events and accurately extracts relevant vehicle information, significantly reducing manual effort in surveillance analysis. The system provides a practical and scalable solution for traffic monitoring, law enforcement, and urban safety applications.