😴 Driver Drowsiness Detection (No YOLO)

A real-time driver drowsiness detection system built using OpenCV + MediaPipe, without using YOLO or any deep-learning object detector.

This project focuses on understanding AI system design, combining computer vision, geometry, and time-based logic instead of training heavy models.

---

📌 Project Objective

To detect driver drowsiness by monitoring eye closure over time using facial landmarks.

Core rule:
If the driver's eyes remain closed for more than a fixed duration → trigger an alert.

---

🧠 Why This Project Is Important

This project teaches real-world AI concepts:

• AI does NOT always mean deep learning
• Feature engineering can replace training
• Geometry + logic = reliable systems
• Time-based reasoning avoids false alarms
• Explainable AI (easy to understand & debug)

This approach is commonly used in:

• Automotive safety systems
• Fleet monitoring
• Driver assistance systems

---

🚀 Features

✅ Real-time webcam processing
✅ Face and eye landmark detection
✅ Eye Aspect Ratio (EAR) calculation
✅ Time-based drowsiness detection
✅ Visual alert on screen
✅ Optional sound alert (macOS)
✅ No dataset required
✅ No model training required
✅ Runs fully offline on CPU

---

🛠️ Technology Stack

• Python 3.9+
• OpenCV – video capture & visualization
• MediaPipe Face Mesh – facial landmarks
• NumPy – numerical operations
• SciPy – distance calculations
• Time module – duration tracking

❌ No YOLO
❌ No deep learning training
❌ No GPU required

---

🧩 How It Works (High-Level)

Webcam Frame
   ↓
Face Detection (MediaPipe)
   ↓
Eye Landmark Extraction
   ↓
Eye Aspect Ratio (EAR)
   ↓
Time-based Logic
   ↓
DROWSINESS ALERT

---

🔢 Core Concept – Eye Aspect Ratio (EAR)

EAR measures how open an eye is using landmark distances.

EAR = (vertical distances) / (horizontal distance)

• Eye open → EAR ≈ 0.30
• Eye closed → EAR ≈ 0.15

If EAR drops below a threshold for enough time, drowsiness is detected.

---

📁 Project Structure

drowny/
├── drowsiness_detection.py   # Main application file
├── venv/                     # Python virtual environment
└── README.md                 # Project documentation

---

⚙️ Installation & Setup

1️⃣ Create virtual environment

python3 -m venv venv
source venv/bin/activate

2️⃣ Install dependencies

pip install opencv-python mediapipe numpy scipy

⚠️ Recommended MediaPipe version:

pip install mediapipe==0.10.8

---

▶️ Run the Project

python3 drowsiness_detection.py

Controls

• Press q to quit

---

🎯 Configuration Parameters

Inside drowsiness_detection.py:

EAR_THRESHOLD = 0.25   # Eye closed threshold
DROWSY_TIME = 2.0      # Seconds eyes must stay closed

Tuning tips:

• Lower EAR_THRESHOLD → more sensitive
• Lower DROWSY_TIME → faster alerts

---

🧠 Key Learnings from This Project

• Blinking ≠ drowsiness
• Time-based logic is critical
• Geometry can replace deep learning
• AI systems need rules, not just predictions
• Debugging environments is part of AI work

---

🚫 Limitations

• Works best for frontal faces
• Single-person detection
• Glasses may affect accuracy
• No head-pose compensation (future improvement)

---

🔮 Possible Enhancements

• Sound alarm using audio file
• Head pose estimation
• Multiple face handling
• Event logging to file
• Mobile / edge deployment
• Combine with deep learning later

---

🧪 Real-World Applications

• Driver safety systems
• Fleet monitoring
• Long-haul transport safety
• Automotive research
• AI learning & demos

---

📜 License

This project is for educational and learning purposes. Use responsibly in real-world safety systems with proper testing.

---

🙌 Final Note

This project proves that:

Good AI is about understanding the problem, not just using big models.

You now understand AI system design, not just AI tools.

---

Happy coding! 🚗💤