A Hybrid YOLO and Centroid-Based Technique for Single-Camera Vehicle Speed Detection and Tracking
Pankaj Kumar Gautam, Sanjeev Kumar · Procedia Computer Science · 2026
Effective vehicle tracking and speed measurement are crucial for managing traffic and enhancing Intelligent Transportation Systems (ITS). This study incorporates vehicle tracking and speed identification from lane-side cameras, employing a bounding box centroid-based technique integrated with the YOLOv9 object detection architecture. Validated on the benchmark dataset VS13, our approach achieved significant improvement in tracking accuracy and speed calculation in traffic videos over existing methods. Testing across various car models in India revealed significant improvement over existing tracking methods in real-time scenarios. Leveraging YOLO based vehicle detection and and incorporated DeepSORT real time tracking incorporating centroid of bounding -box demonstrates reliablity in detection on side angle tracking and detection in vaious traffic scenarios on numerous vehicle model. Our method operates efficiently at 30 FPS with low computational intensity, showcasing the capability of YOLOv9 for realtime vehicle tracking applications while offering a strong foundation for future advancements in traffic management and safety protocols.