Drowsiness Detection System

Overview

The Drowsiness Detection System is a project designed to monitor a person's alertness in real-time by analyzing facial features.
By utilizing computer vision and machine learning techniques, the system aims to detect signs of drowsiness and provide timely alerts — particularly useful for applications like driver monitoring.

This repository focuses on illustrating the full development process, including data capture, auto-labeling, model training, and detection pipeline integration.

---

Features

• Real-time Monitoring: Detects signs of drowsiness using a webcam or video input.
• Dual Model Detection: Separate YOLOv8 models for eye closure detection and yawning detection.
• Facial Landmarks Analysis: Tracks eye status, head position, and mouth movements.
• Data Capture Pipeline: Tools to collect and organize custom datasets.
• Auto Labeling with GroundingDINO: Automated bounding box generation.
• User Interface: Built with PyQt5 for real-time visualization and alerts.

---

Key Files

• AutoLabelling.py: Script for automated bounding box labeling using GroundingDINO.
• CaptureData.py: Records and logs video data for analysis or training.
• DrowsinessDetector.py: Core detection script integrating real-time inference and alerts.
• LoadData.ipynb: Loads and preprocesses datasets.
• RedirectData.ipynb: Organizes and redirects captured data for training.
• train.ipynb: Notebook for training the YOLO models.

---

Installation

1. Clone the repository:

   git clone https://github.com/tyrerodr/Real_time_drowsy_driving_detection.git
   cd Real_time_drowsy_driving_detection

2. Create a virtual environment:

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Run the detection system:

   python DrowsinessDetector.py

---

Usage

• Real-Time Detection: Run DrowsinessDetector.py with a connected webcam to monitor drowsiness.
• Data Capture: Use CaptureData.py to collect video frames for training or testing.
• Training New Models: Use train.ipynb to retrain the models on your custom datasets.

---

How It Works

The system uses two separate YOLOv8 models:

1. Eye Detection Model:

   • Classifies eyes as open or closed.
   • Trained on public datasets:
     • Eyes Dataset
     • MRL Eye Dataset
   • ~53,000 images for training, ~3,000 images for validation.

2. Yawning Detection Model:

   • Detects yawning (mouth open) vs not yawning (mouth closed).
   • Trained on:
     • Yawning Dataset

Auto Labeling:
GroundingDINO was used to generate bounding boxes for YOLO training to improve dataset quality.

Once trained, the models' predictions are combined with confidence thresholds and visualized in a PyQt5 GUI.

---

Technologies Used

• Python
• YOLOv8 – Object detection framework.
• OpenCV – Computer vision tasks.
• GroundingDINO – Auto-labeling tool.
• TensorFlow / Keras – Model training.
• PyQt5 – Graphical user interface.

---

Important Note

This repository is intended primarily to showcase the development process of a drowsiness detection system — including data collection, model training, and real-time integration.

The uploaded model weights are preliminary and not fully trained to convergence.
They are mainly for demonstration purposes, and final production-ready models are maintained separately.

We appreciate any feedback and contributions to improve the system.

---

Future Improvements

• Integration with Wearables: Add heart rate or other vitals monitoring.
• Multi-Person Detection: Extend detection to multiple subjects simultaneously.
• Mobile Deployment: Create a mobile app version for real-time on-the-go monitoring.

---

Eng. Tyrone Eduardo Rodriguez Motato
Computer Vision Engineer
Guayaquil, Ecuador
Email: tyrerodr@hotmail.com