Collision Alert System: Computer Vision for Vehicle Safety

This paper introduces an innovative, vision-based Collision Alert System (CAS) aimed at significantly enhancing vehicle safety in complex driving environments. Our approach advances beyond existing YOLO-based Advanced Driver- Assistance Systems (ADAS) through its unique integration of YOLOv8’s real-time object detection with specialized Dynamic Region of Interest (ROI) processing and Angular Methods for precise trajectory analysis. This framework, combined with robust object tracking, enables proactive collision prediction. We detail the system’s architecture and operational flow, emphasizing its real-time performance. Rigorous evaluation confirms a True Positive Rate exceeding 95% and a remarkably low False Positive Rate under 2%, while maintaining over 30 Frames Per Second (FPS) for timely driver alerts. The paper also discusses core algorithmic principles and effective strategies for addressing real- world challenges like variable illumination and camera instability. A video demonstrating the system’s live operation is available online. Future work focuses on multi-camera data fusion and integration with active vehicle control.