DriverCo is an AI-powered, real-time drowsiness detection and voice assistant system designed to make driving safer. Built at HackUPC 2025, this project combines computer vision, generative AI, and voice technologies to detect early signs of fatigue and engage drivers through natural conversation.

• 👁️ Real-time Drowsiness Detection using fine-tuned YOLOv8 on eye and mouth cues (e.g. eye closure, yawning)
• 🧠 Conversational Assistant powered by Google Gemini API
• 🎙️ Speech-to-Text & Text-to-Speech interface for hands-free interaction
• 📊 Custom Trained Model on publicly available fatigue datasets (50 epochs)
• 🎥 Live Webcam Inference for testing and real-time simulation

DriverCo-HackUPC/
│
├── driverco-model-training.ipynb     # Notebook for training the YOLOv8 model
├── DriverCo_VF.ipynb                 # Final system integrating detection + Gemini AI
├── README.md                         # You're here

git clone https://github.com/yourusername/DriverCo-HackUPC.git
cd DriverCo-HackUPC

We recommend running on Google Colab. The main packages include:

• ultralytics
• opencv-python
• google-generativeai
• edge-tts
• Pillow, numpy, matplotlib

Open driverco-model-training.ipynb to:

• Visualize dataset
• Fine-tune YOLOv8
• Save your custom model

Note: We trained our model on a custom dataset for 50 epochs to optimize accuracy for drowsiness signs.

Open DriverCo_VF.ipynb to:

• Run real-time webcam detection
• Trigger conversational assistant when drowsiness is detected

We used a publicly available dataset focused on drowsy driving signs. It includes labeled facial landmarks for eye and mouth states. The model was trained in YOLO format for compatibility.

• YOLOv8 — Object detection
• Google Gemini API — Conversational GenAI
• OpenCV — Real-time computer vision
• Edge-TTS — Speech synthesis
• Google Colab — Development environment

Made with 💡 at HackUPC 2025 by:

• Kamel Yamani
• Marwa Nair

This project is open-source and available under the MIT License.