🚗 DriveEye

AI-Powered Driver Monitoring System

Real-time drowsiness detection using YOLOv5 + TensorRT, built for public transport safety.

---

✨ Features

• 🧠 Binary YOLOv5 model (drowsy / not_drowsy)
• 🔁 3-second buffer detection logic
• ⚡ TensorRT optimized for Nvidia GPUs
• 📷 Live USB/CSI camera feed
• 📝 Incident logging with timestamped frame & metadata
• 🧪 High-confidence data collection for retraining

---

1. 🐳 Pull Docker Image

sudo docker pull {pytorch image with tensorrt and cuda}

---

📦 Project Setup

Clone Repository

git clone https://github.com/your-username/DriveEye.git
cd DriveEye

Dataset (Roboflow)

Use Driver Drowsiness v3 in YOLOv5 format:

Dataset1/
├── data.yaml
├── train/
│   ├── images/
│   └── labels/
└── valid/
    ├── images/
    └── labels/

---

🏋️ Model Training

python3 Training/train.py

• Output: Training/runs/detect/weights/best.pt

---

🧪 Testing the Model

python3 Training/test.py

---

🧠 Export to TensorRT

python3 Models/tensorrt_export.py

• Outputs: Models/best.engine

---

🎥 Run Inference

Update camera index if needed in Inference/video_infer.py:

cap = cv2.VideoCapture(0)  # Replace 0 if needed

Then run:

python3 main.py --mode inference

---

📂 Directory Structure

DriveEye/
├── Dataset1/
├── Inference/
├── Models/
├── Training/
├── Logger/
├── DataCollector/
├── logs/
│   ├── incidents/
│   └── high_confidence/
├── Dockerfile
├── main.py
├── requirements.txt
└── README.md

---

🐳 Docker

sudo docker build -t driveeye .
sudo docker run -it --runtime nvidia --network host \
  -v $PWD:/app --workdir /app driveeye \
  python3 main.py --mode inference

---

🧩 Troubleshooting

• 📷 Camera not detected? Try another index:

v4l2-ctl --list-devices

---

📌 Roadmap

• ✅ GPS integration
• 📈 MLflow experiment tracking
• ♻️ Retraining pipeline w/ real-world data
• 🚦 Multi-behavior detection (texting, yawning, etc.)

---

🤝 Contributing

PRs and issue reports are welcome!

---