VIETNAMESE SIGN LANGUAGE RECOGNITION

A system for recognizing Vietnamese Sign Language using deep learning and computer vision techniques, tailored specifically for Vietnamese sign language.

Demo

Table of Contents

• Overview
• Features
• Requirements
• Installation
• Usage
  • Running the Application
  • Training from Scratch

Overview

The Vietnamese Sign Language Recognition system leverages deep learning models and computer vision to interpret Vietnamese sign language gestures. It uses MediaPipe for landmark detection, TensorFlow for model training, and Streamlit for a user-friendly interface. The system supports recognition through video files or live webcam feeds.

Features

• Automated Video Download: Automatically downloads videos for training data.
• Data Preprocessing: Processes and augments data for model training.
• Sign Language Recognition: Recognizes Vietnamese sign language gestures via video or webcam input.
• User Interface: Provides a Streamlit-based web interface for easy interaction.

Requirements

• Software:
  • Python 3.8 or higher
  • TensorFlow 2.x
  • Scikit-learn
  • MediaPipe
  • OpenCV
  • Streamlit
• Hardware:
  • Webcam (required for webcam recognition)
  • GPU (recommended for model training)

Installation

1. Clone the repository

git clone https://github.com/baamvu/quoxbau

cd Vietnamese-Sign-Language-Recognition

Alternatively, download the ZIP file from GitHub and extract it.

2. Install Dependencies

Ensure Python is installed. If not, you can download and install it from the official Python website. Then, install the required libraries:

pip install -r requirements.txt

Usage

The system can be used either by running the pre-trained model or by training a new model from scratch.

Running the Application

To use the pre-trained model with the Streamlit interface:

streamlit run main.py

This launches a web interface where you can upload videos or use a webcam for sign language recognition.

Training from Scratch

To train a new model, follow these steps:

1. Clear Previous Data (optional).

Get-ChildItem -Path "./" -Directory | Remove-Item -Recurse -Force

2. Download Training Data.

python download_data.py

3. Process Data.

python create_data_augment.py

4. Train the Model.

• Open training.ipynb in a Jupyter Notebook environment.
• Run all cells to train the model.
• Note: Training is computationally intensive and best performed on a GPU-enabled device.

5. Run the Application.

streamlit run main.py