AI-Based Green Light Optimization using Computer Vision

Urban traffic systems still rely largely on fixed timer traffic lights. These timers do not adapt to real-time traffic conditions, which often leads to congestion, unnecessary waiting time, and increased fuel consumption. To explore a more intelligent approach, I built the Metropolitan AI Control Center, a traffic signal optimization system that combines Deep Reinforcement Learning, Computer Vision, and traffic simulation. The goal of the project is to replace static traffic signal timers with an AI agent that continuously learns how to manage intersections based on real-time traffic conditions. Project Overview The system operates on a simulated 10-intersection city grid using Eclipse SUMO (Simulation of Urban Mobility). A Deep Q-Network (DQN) agent learns how to control signal phases in order to reduce overall waiting time and prevent congestion from spreading across intersections. Vehicle density is estimated using YOLOv8-based computer vision, while the entire system is monitored t