Optimizing Traffic Light Control using YOLOv8 for Real-Time Vehicle Detection and Traffic Density

The problem of traffic congestion becomes crucial as the number of vehicles on the roads, especially in big cities, continues to rise. An effective and cost-efficient solution is managing traffic lights based on density. This paper introduces a method utilizing deep neural networks to control traffic lights by analyzing surveillance camera footage and adjusting the lights automatically. Specifically, we employ the state-of-the-art YOLOv8 model to identify vehicles on the road from camera images, assess traffic density, and optimize the timing of traffic lights accordingly. The evaluation results show that the proposed model demonstrates a high accuracy in vehicle detection, achieving a mAP50 of up to $\mathbf{9 7 \%}$. In addition, the results of vehicle counting and traffic light control are also tested in various contexts.