👁️ EyeQAI — Real-Time Drowsiness & Attention Detection System

EyeQAI is a real-time AI-based monitoring system that detects user alertness using computer vision and deep learning techniques.

It classifies user state into:

• Focused — eyes open, head forward
• Drowsy — prolonged eye closure
• Inattentive — head turned away

---

🚀 Key Features

• Hybrid AI detection (CNN + EAR + Head Pose)
• Real-time webcam monitoring
• Frame smoothing to reduce false positives
• Attention score calculation
• REST API for metrics and logs

---

🧠 Architecture Overview

EyeQAI combines three signals for robust detection:

• CNN Model (ResNet50) → Eye state classification
• EAR (Eye Aspect Ratio) → Detects eye closure
• Head Pose (Yaw Angle) → Detects distraction

Final output is stabilized using frame smoothing (3 consecutive frames) to avoid sudden spikes.

---

📂 Project Structure

EyeQAI/
└── app/
    ├── app.py
    ├── detection.py
    ├── train_model.py
    ├── utils.py
    ├── requirements.txt
    ├── templates/
    ├── static/

⚠️ Note:

• models/, data/, and venv/ are excluded from GitHub due to size limitations.

---

⚙️ Setup Instructions

1. Create Virtual Environment

cd app
python -m venv venv
venv\Scripts\activate     # Mac/Linux: source venv/bin/activate
pip install -r requirements.txt

---

2. Train the Model

python train_model.py

This will generate the trained model inside the models/ folder.

---

3. Run the Application

python app.py

Open in browser: 👉 http://localhost:5000

---

🌐 Deployment (Render)

• Root Directory → app
• Build Command

pip install -r requirements.txt

• Start Command

python app.py

⚠️ Render automatically manages the virtual environment. No need to activate venv.

---

📊 API Endpoints

Endpoint	Description
/video_feed	Live webcam stream
/metrics	Real-time detection data
/logs	Event history
/summary	Session summary
/health	System health

---

🎯 Detection Logic

Condition	Output
EAR < 0.22	Drowsy
Yaw > 25°	Inattentive
Otherwise	Focused

Frame smoothing ensures stable predictions across multiple frames.

---

⚠️ Important Notes

• Large files (models, datasets) are excluded using .gitignore
• Model can be stored externally (Google Drive / cloud)
• System can run in EAR-only mode if model is not available

---

🛠️ Tech Stack

• Python
• OpenCV
• MediaPipe
• PyTorch / TensorFlow
• Flask

---

📌 Future Improvements

• Cloud-based model loading
• Multi-user analytics dashboard
• Mobile deployment