SignFlow | Real-Time Sign Language Alphabet Recognition System

Overview

This project is a real-time system for recognizing English Sign Language letters. Developed at Doğuş University by the Software Engineering team, it uses machine learning and computer vision to help hearing-impaired individuals communicate more easily and support sign language learning.

The system recognizes letters with over 90% accuracy and handles dynamic gestures like "J" and "Z". It is designed to be simple, reliable, and accessible to users of all ages.

Project Structure

SignFlow/
├── Data/ - Dataset
├── gui.py - GUI application
├── egitim.py - Training script
├── import os.py - Helper script
├── test.py - Test script
├── test_model.py - Model testing
├── train_model.py - Model training
├── requirements.txt
├── README.md
├── .gitignore

Key Features

• Real-Time Recognition: Detects hand gestures instantly using a camera.
• High Accuracy: Recognition accuracy exceeds 90%.
• Dynamic Gesture Support: Handles complex gestures like "J" and "Z".
• User-Friendly: Simple interface suitable for everyone.
• Inclusive Impact: Helps raise awareness of sign language and accessibility.

Goals

• Enable hearing-impaired individuals to communicate more easily.
• Provide an educational tool for learning sign language.
• Promote societal awareness and inclusivity.

Technical Details

• Framework: TensorFlow
• Language: Python
• Libraries: OpenCV, Numpy
• Model: Convolutional Neural Network (CNN)
• Dataset: Custom dataset with static and dynamic gestures

How It Works

1. Input: Capture hand gestures via camera.
2. Preprocessing: Resize, normalize, and prepare images for recognition.
3. Prediction: CNN processes images and identifies letters.
4. Output: Display letters in real-time, optionally with voice feedback.

Installation

git clone https://github.com/YourUsername/SignFlow.git
cd SignFlow
pip install -r requirements.txt
python main.py

Performance

• Accuracy: >90% on training and validation datasets.
• Dynamic Gestures: Successfully recognizes "J" and "Z".
• Low-End Device Support: Runs efficiently on basic hardware.
• User Feedback: Simple, accurate, and practical.

Contributors

• Ayşe Ceren Doğan
• Cemre Dağ
• Emir Ekrem Kaya
• Hatice Uçar

Note

The project name is intended for future research references and was not used during the original research project.