Static, pre-programmed traffic signals are fundamentally inefficient. They cannot adapt to:
-
Sudden traffic surges during peak hours or events.
-
The urgent need of an ambulance or fire truck.
-
Uneven traffic flow, where one lane is congested while another is empty.
-
Pedestrian crossing needs.
"Our solution provides the blueprint for integrating with existing government camera infrastructure to deliver real-time, adaptive traffic management and instant responses to critical road events."
We've engineered a multi-layered AI system where each component has a distinct and critical role, mimicking a human-like decision-making process.
A state-of-the-art YOLOv8 Computer Vision model acts as our eyes on the street. It processes live video feeds to provide a constant stream of high-fidelity data:
-
Detects & Classifies: Accurately identifies cars, buses, trucks, motorcycles, and pedestrians.
-
Counts & Tracks: Maintains a real-time count for each lane.
-
Calculates Density: Measures queue lengths to understand congestion levels.
The raw data is fed to our Q-Learning Reinforcement Learning agent. This is the strategic brain of the operation. It has been trained over thousands of simulations to learn the optimal traffic signal strategy. Its single goal: minimize wait times and maximize traffic throughput. It constantly asks, "What's the most efficient signal phase to activate right now?"
AI needs oversight. A powerful Rule-Based Optimization Engine acts as a guardian, supervising the Q-Learning agent's decisions to ensure safety, fairness, and responsiveness to critical events. This layer enforces non-negotiable rules:
-
Emergency Vehicle Preemption: Detects emergency vehicles and initiates a high-priority override sequence to clear a path, saving precious seconds.
-
Gridlock Prevention: If a queue length exceeds a critical threshold, it triggers an override to service that lane immediately.
-
Starvation Prevention: A fairness timer ensures no vehicle waits indefinitely, guaranteeing a minimum level of service for all lanes.
| Innovation | Description & Impact |
|---|---|
| Adaptive Signal Control | Green light duration is allocated based on real-world demand, not a fixed clock. Impact: Drastically reduces average wait times and vehicle idling. |
| Emergency Preemption | A professional-grade clearing sequence (yellow -> all-red -> dedicated green) provides a clear path for first responders. Impact: Saves lives by shortening emergency response times. |
| Live Digital Twin Dashboard | A real-time React dashboard visualizes every detected vehicle, AI decision, and signal state. Impact: Provides complete transparency and operational oversight for city traffic managers. |
| Guaranteed Fairness | Our “starvation” prevention algorithm ensures that low-traffic lanes aren’t ignored, promoting equitable flow. Impact: Increases driver satisfaction and prevents isolated congestion. |
- AI & Computer Vision: Python, OpenCV, PyTorch, YOLOv8, Q-Learning
- Backend & Data Pipeline: Python, FastAPI, WebSockets
- Frontend & Visualization: React.js, TypeScript, Vite, TailwindCSS, Recharts
- Deployment: Vercel (Frontend), Railway (Backend)
- Python 3.10+
- Node.js and npm/yarn
- A video file (
my_video.mp4) of traffic for the AI to process
cd backend
python -m venv venv
source venv/bin/activate (for MacOS) || venv\Scripts\activate (for Windows)
pip install -r requirements.txt
uvicorn main:app --reload
python run_live_agent.pycd frontend
npm install
npm run dev