Urban traffic congestion remains a critical challenge, demanding intelligent and adaptive solutions for efficient management. This paper introduces a dynamic traffic signal control system powered by the You Only Look Once version 8 (YOLOv8) model trained on KITTI dataset for real-time vehicle detection and traffic optimization. Leveraging YOLOv8’s advanced object detection capabilities, the system accurately detects and counts vehicles at intersections, dynamically adjusting traffic light durations to improve flow and reduce congestion. The system operates in three modes: normal, ensuring balanced traffic flow across lanes; priority, minimizing delays for high-density lanes; and emergency, prioritizing uninterrupted passage for emergency vehicles. The evaluation shows a 95% reduction in Average Waiting Time and a 75% improvement in Throughput under heavy traffic scenarios. Although the framework is scalable and made to integrate with current infrastructure, issues like camera calibration and environmental unpredictability may arise during real-world implementation. In smart city settings, the suggested approach provides a reliable way to improve urban mobility by fusing deep learning with adaptive traffic control techniques.