🚗 Drowsiness Control System for Autoware

A real-time driver monitoring and safety system built on ROS2 that detects drowsiness using computer vision and progressively applies braking via Autoware.

---

✨ Features

• 👁️ Real-time drowsiness detection using facial landmarks
• 🧠 Multi-signal analysis (EAR, MAR, head pose)
• ⚙️ 6-stage braking state machine
• 🚦 Seamless Autoware integration
• 📊 Live debug visualization & status topics
• 🛡️ Safety-first gradual intervention (no sudden braking)
• ✅ Fully tested (16/16 tests passing)

---

🧠 System Overview

graph LR
    CAM[Camera] --> DET[Detector]
    DET --> CTRL[Braking Controller]
    CTRL --> AUTO[Autoware Interface]
    AUTO --> VEH[Vehicle Control]

Loading

---

📁 Project Structure

drowsiness_control/
├── config/                     # Runtime parameters
├── launch/                     # ROS2 launch files
├── drowsiness_control/
│   ├── camera_input.py
│   ├── drowsiness_detector.py
│   ├── braking_controller.py
│   ├── autoware_interface.py
│   ├── drowsiness_node.py
│   └── constants.py
└── test/                       # Unit tests (16 total)

---

⚙️ Installation

1. Clone into Autoware workspace

cd ~/project/autoware/src
git clone <your-repo-url>

2. Install dependencies

pip3 install mediapipe opencv-python

3. Build

cd ~/project/autoware
colcon build --packages-select drowsiness_control --symlink-install
source install/setup.bash

---

▶️ Usage

Run the system

ros2 launch drowsiness_control drowsiness_control.launch.py

Run with custom parameters

ros2 launch drowsiness_control drowsiness_control.launch.py \
    camera_index:=0 \
    max_velocity_mps:=8.33 \
    publish_debug:=true

---

📡 ROS2 Topics

Topic	Description
/drowsiness_control/status	Drowsiness level output
/drowsiness_control/braking_state	Current braking state
/drowsiness_control/debug_image	Debug visualization
/planning/scenario_planning/max_velocity	Velocity control

---

🧠 Detection Method

The system evaluates driver alertness using:

• EAR (Eye Aspect Ratio) → detects eye closure
• MAR (Mouth Aspect Ratio) → detects yawning
• Head Pose Estimation → detects nodding

These signals are combined into a normalized score (0–1) and mapped to:

• ALERT
• MILDLY_DROWSY
• DROWSY
• SEVERELY_DROWSY

---

🚦 Braking Logic

State	Behavior
NORMAL	No action
ALERT_PHASE	Warning + 80% speed
GENTLE_BRAKE	Gradual slowdown
HARD_BRAKE	Full stop
EMERGENCY_STOP	Emergency command
RECOVERY	Smooth speed restoration

---

🛡️ Safety Design

• ❌ No sudden braking
• 📉 Gradual velocity reduction using jerk limits
• ⏱️ Progressive escalation (~25 seconds)
• 🔄 Controlled recovery (~8 seconds ramp-up)
• 🚫 No false positives if camera fails

---

🧪 Testing

pytest

✔️ Result:

16 passed in 0.75s

---

📊 Configuration

Edit parameters in:

config/drowsiness_params.yaml

Example:

ear_threshold: 0.21
mar_threshold: 0.75
velocity_steps: [0.80, 0.60, 0.40, 0.20, 0.10, 0.0]

---

🔍 Monitoring & Debugging

# View drowsiness state
ros2 topic echo /drowsiness_control/status

# View braking state
ros2 topic echo /drowsiness_control/braking_state

# View debug feed
ros2 run rqt_image_view rqt_image_view /drowsiness_control/debug_image

---

🚀 Future Work

• Night-time robustness
• Driver-specific calibration
• Multi-camera fusion
• ML-based fatigue prediction
• Integration with cabin sensors

---

🤝 Contributing

Contributions are welcome!

1. Fork the repo
2. Create a feature branch
3. Commit changes
4. Open a pull request

---

📜 License

This project is licensed under the MIT License.

---

👤 Author

Abinesh K

---

⭐ Show Your Support

If you find this project useful:

• ⭐ Star the repo
• 🍴 Fork it
• 🧑‍💻 Contribute

---

🧩 Tech Stack

• ROS2
• Python
• OpenCV
• MediaPipe
• Autoware

---

📌 Summary

A complete end-to-end safety system combining:

• Computer Vision
• ROS2 Middleware
• Autonomous Vehicle Control
• Real-time State Machines

Designed for real-world autonomous driving safety applications.

---