Real-Time Traffic Monitoring Using Deep Learning and Multi-Object Tracking

Accurate traffic monitoring traditionally relies on physical sensors such as inductive loops or radar, which are costly to deploy and maintain. This paper presents a lightweight, camera-based alternative using modern computer vision. A YOLOv8 object detector is combined with Kalman filter–based multi-object tracking to estimate vehicle trajectories, frame-level density, and a normalized congestion score from a single fixed camera. Experiments on a two-minute highway video demonstrate that the system produces stable density curves and congestion profiles, suggesting that commodity cameras and deep learning can serve as a practical, scalable alternative to traditional traffic sensors. All code is publicly available at https://github.com/bforce541/traffic-density-yolo-kf