🛌 AI Drowsiness & Yawning Detection System with Telegram Alerts

This project is a complete real-time drowsiness and yawn detection system built using OpenCV, MediaPipe, Tkinter GUI, and Telegram Bot for alert notifications. It includes:

• Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) calculations
• TTS voice alerts
• Pre- and post-alert video recording
• Telegram alerts with 10s video
• CUDA GPU acceleration (if available)

---

💾 Features

✅ Real-time face mesh detection using MediaPipe ✅ EAR/MAR calculation for eye closure and yawning ✅ Auto video recording (5s before + 5s after alert) ✅ Telegram bot alert with video proof ✅ Tkinter-based GUI with live feed ✅ Text-to-speech (TTS) alerts ✅ GPU acceleration using OpenCV CUDA

---

📁 Project Structure

.
├── drowsy_detector.py       # Main source code (uploaded by user)
├── README.md                # This file
└── /drowsy_video/           # Auto-saved alert videos

---

🧰 Requirements

Install all Python dependencies using:

pip install opencv-python mediapipe pyttsx3 python-telegram-bot Pillow

✅ Optional for GUI support:

pip install tkinter

✅ Optional for GPU acceleration:

pip install opencv-contrib-python

---

⚙️ Configuration

Open the drowsy_detector.py file and update the following:

TELEGRAM_TOKEN = '<YOUR_BOT_TOKEN>'
CHAT_ID = '<YOUR_TELEGRAM_CHAT_ID>'

If not using Telegram alerts, just leave TELEGRAM_TOKEN = ''.

✅ Ensure the following directory exists:

mkdir -p /home/pi/drowsy_video

---

🚀 How to Run

On any PC with Python:

python drowsy_detector.py

You’ll see a GUI window with:

• EAR/MAR values
• Start/Stop buttons
• Live video feed

---

📆 Features in Action

Feature	Description
🟢 Drowsy Detection	Alerts if eyes closed for EYE_FRAMES
🟡 Yawn Detection	Alerts if mouth open for MAR_FRAMES
🎥 Video Recording	5 seconds before & after alert
🗣️ TTS	Audible alert
📤 Telegram Bot	Sends video to your chat

---

💡 Thresholds (tweak as needed)

EYE_THRESH = 0.28        # Eye Aspect Ratio
EYE_FRAMES = 51          # 1.7 seconds at 30 FPS

MAR_THRESH = 0.6         # Mouth Aspect Ratio
MAR_FRAMES = 60          # 2 seconds at 30 FPS

---

⚡ GPU Acceleration

If you have an NVIDIA GPU:

• Uses cv2.cuda to process frames
• Set CUDA device using cv2.cuda.setDevice(0)
• Automatically falls back to CPU if CUDA not found

---

🧟 Telegram Bot Setup (Optional)

1. Create bot using @BotFather
2. Note down the BOT TOKEN
3. Get your chat_id using this method

---

🚫 Exit & Cleanup

• Clicking Exit or closing the GUI will:

  • Release camera
  • Stop TTS
  • Finalize video
  • Exit gracefully

---

🧠 Author

Created by Deventhira Kesavaraj BE.EEE, KPRIET Project: AI Drowsy Detection System with Telegram Alert Support

---

📸 Sample Alert Message

🛌 ALERT: DROWSINESS DETECTED!
Capturing 10-second video...

🔗 Then sends the video via Telegram!