🤟 Indian Sign Language Recognition System

An AI-powered real-time Indian Sign Language (ISL) recognition and translation system.
Includes data collection tools, training notebooks, and a real-time translation app with speech synthesis.

---

📑 Table of Contents

• Overview
• Features
• Project Structure
• Installation
• Usage
  • Real-time Translation
  • Data Collection
• Model Training
• Technical Details
• Requirements
• Contributing
• License
• Acknowledgments
• Future Enhancements
• Support

---

🔎 Overview

This project enables real-time ISL-to-text and speech translation.
It supports 35 sign classes (A–Z, 0–9, space, delete) and comes with tools for dataset collection and deep learning model training using MobileNetV2 and ResNet.

---

✨ Features

• 🖐️ Hand Detection – MediaPipe for accurate landmark detection
• 🧾 Multi-class Recognition – 35 classes supported
• 🗣️ Live Translation – Sign-to-text with speech synthesis
• 💻 Modern GUI – Clean, dual camera view interface
• 📊 Data Collection Tools – Automated dataset generation
• 🧠 Multiple Models – MobileNetV2 & ResNet architectures
• ⚡ Performance Optimized – GPU acceleration, efficient inference

---

📁 Project Structure

project/
├── Script.py                           # Real-time translation app
├── collect_imgs.py                     # Data collection script
├── Indian_sign_language_Mobilenetv2.ipynb   # MobileNetV2 training
├── Indian sign language resnet.ipynb        # ResNet training
├── indian-sign-language-classification.ipynb # Classification notebook
├── data_images/                        # Image dataset (generated)
└── README.md                           # Documentation

---

🚀 Installation

Prerequisites

• Python 3.8+
• Webcam
• CUDA-compatible GPU (optional, for faster training)

Setup

git clone <repository-url>
cd indian-sign-language-recognition
pip install -r requirements.txt
pip install opencv-python mediapipe tensorflow pillow pyttsx3

---

📖 Usage

Real-time Translation

python Script.py

Controls:

• Ctrl+S → Speak detected text
• Ctrl+C → Clear text area
• Ctrl+P → Pause/resume detection
• Esc → Stop speech

Data Collection

python collect_imgs.py

Steps:

1. Run the script
2. Select class (A–Z, 0–9, space, delete)
3. Press Q to start capturing (500 images/class)
4. Press Esc to quit early

---

🧠 Model Training

• MobileNetV2 → Transfer learning, augmentation, early stopping
• ResNet → Pre-trained architecture with fine-tuning

Training Highlights:

• Input: 224×224 RGB images
• Optimizer: Adam, Loss: Categorical Crossentropy
• Callbacks: Early stopping, LR scheduler
• Epochs: 50 (with early stopping)

---

🔧 Technical Details

• Hand Detection: MediaPipe (21 landmarks/hand)
• Model Input: 224×224×3 images
• Output: 35-class softmax prediction
• Optimizations: GPU acceleration, memory management, threaded speech synthesis

---

📋 Requirements

Core:

tensorflow>=2.8.0
opencv-python>=4.5.0
mediapipe>=0.8.0
numpy>=1.21.0
pillow>=8.0.0
pyttsx3>=2.90
tkinter (built-in)

Optional:

matplotlib>=3.5.0
scikit-learn>=1.0.0
pandas>=1.3.0

System:

• OS: Windows / macOS / Linux
• RAM: 8GB+ (16GB recommended)
• GPU: NVIDIA CUDA (optional)

---

🤝 Contributing

1. Fork repo
2. Create branch (feature/xyz)
3. Commit changes
4. Push & open Pull Request

---

📄 License

MIT License – see LICENSE.

---

🙏 Acknowledgments

• MediaPipe – hand detection
• TensorFlow – deep learning
• OpenCV – computer vision
• Indian Sign Language community – inspiration & feedback

---

🔮 Future Enhancements

• Dynamic sign (words/phrases) support
• Multi-hand gesture recognition
• Real-time video translation
• Mobile app version
• Cloud-based processing
• Communication platform integration

---

📞 Support

• Open a new issue with details (system specs + error logs)

---

✨ Made with ❤️ for the Indian Sign Language community