Advancing Urban Traffic Control with IoT and Deep Learning: A YOLOv8 and LSTM-based Adaptive Signal System

Urban traffic congestion remains a significant challenge due to increasing vehicle density, inefficient signal control, and unpredictable traffic patterns. Traditional fixed-time signal systems fail to adapt dynamically to varying congestion levels, leading to increased delays, fuel consumption, and environmental pollution. To address these limitations, this research proposes an IoT and deep learning-based adaptive traffic signal system integrating YOLOv8 for real-time vehicle detection and LSTM for congestion prediction. Traffic data is collected using IoT sensors, including cameras, ultrasonic sensors, RFID modules, GPS trackers, and air quality sensors. YOLOv8 processes real-time camera feeds to detect vehicles, classify them based on type, and estimate traffic density. Concurrently, an LSTM-based predictive model analyzes historical traffic patterns and external factors such as peak hours and weather conditions to forecast congestion trends. The system dynamically adjusts traffic signal durations based on congestion probability scores, optimizing road utilization and minimizing vehicle waiting times. Extensive training and evaluation using real-world traffic datasets demonstrate significant improvements in congestion reduction compared to conventional traffic control methods. Performance metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) validate the reliability of the predictive model. Additionally, air quality monitoring highlights the environmental benefits of reduced emissions. This intelligent traffic management framework enhances urban mobility, improves emergency vehicle response times, and contributes to sustainable smart city development.