The Adaptive Traffic Conductor: Deep Reinforcement Learning for Intelligent Signal Coordination

Traffic congestion and emergency response delays remain critical challenges in urban transportation.This project presents an AI-driven traffic management system that combines real-time computer vision and deep reinforcement learning (DRL) to optimize signal control.A YOLOv5 object detector (implemented in PyTorch) processes live video frames to count vehicles per lane and identify emergency vehicles.These counts and emergency flags form the state input to a DRL agent, which selects traffic signal phases on a second-by-second basis, prioritizing congested approaches and clearing paths for ambulances and fire trucks.In simulated tests our adaptive controller significantly reduced average vehicle delay and queues compared to fixed-timing signals, and nearly eliminated wait time for detected emergency vehicles.The system logs per-second metrics (counts, signal states, emergencies) and exports CSV reports for analysis.Our results demonstrate that an RLpowered, vision-based control strategy can enhance traffic flow and emergency access.