Bangladeshi Sign Language Detection

A YOLOv10-based system for recognizing static Bangladeshi Sign Language gestures and translating them into text.

Developed a Bangladeshi Sign Language (BdSL) Detection System to help bridge communication barriers faced by deaf and mute individuals. The system recognizes static Bangladeshi Sign Language hand gestures and translates them into text, supporting more inclusive and accessible communication. The proposed approach leverages YOLOv10, a state-of-the-art object detection model, trained on a custom dataset of labeled BdSL gesture images. The model was designed to provide accurate and efficient recognition of 14 unique static signs.

YOLOv10 Python Computer Vision Deep Learning Object Detection Sign Language Recognition Bangladeshi Sign Language 14 Static Signs Custom Dataset Text Translation

View Code

Project Overview

There is a critical need for accurate Bangladeshi Sign Language (BdSL) detection systems to create a more inclusive environment for people who are deaf and mute. Communication barriers can contribute to social isolation and limit access to education, services, and everyday interactions.

This project proposes a computer vision-based system that recognizes Bangladeshi Sign Language hand gestures and translates them into text. By facilitating communication between sign language users and others, the system aims to reduce communication barriers and support greater social inclusion.

The proposed method uses YOLOv10, a state-of-the-art object detection model, to achieve accurate and efficient BdSL gesture recognition. A custom dataset of 1,949 labeled images covering 14 unique static signs was used to train and evaluate the model.

Methodology

Custom Dataset

A curated dataset of 1,949 labeled images representing 14 unique static Bangladeshi Sign Language gestures.

YOLOv10 Model

A custom YOLOv10 object detection model trained to identify and recognize static BdSL hand gestures.

Model Training

The model was trained on labeled gesture images to improve recognition accuracy and robustness across the selected signs.

Text Translation

Recognized hand gestures are translated into text to support communication between sign language users and others.

System Workflow

1. Input Image

A hand gesture image is provided as input to the detection system.

2. Gesture Detection

The trained YOLOv10 model detects the hand gesture and identifies the corresponding sign class.

3. Sign Recognition

The detected gesture is classified into one of the 14 supported Bangladeshi Sign Language signs.

4. Text Output

The recognized sign is converted into text to facilitate communication.

Dataset and Model Details

Attribute	Details
Project Type	Computer Vision and Sign Language Recognition
Model	YOLOv10
Dataset	Custom labeled Bangladeshi Sign Language dataset
Total Images	1,949
Number of Signs	14 unique static signs
Recognition Type	Static hand gesture detection
Output	Recognized sign translated into text
Primary Objective	Bridging communication barriers through BdSL recognition

Performance Results

The proposed method was evaluated on the custom dataset containing 1,949 images of 14 unique signs. The model achieved the following performance across all classes.

Metric	Performance
F1-Confidence Rate	86%
Recall-Confidence Rate	98%
Precision-Confidence Rate	100%
Precision-Recall Rate	90.3%
Overall Average Accuracy	90.67%

Key Contributions

Inclusive Communication: Develops a system intended to reduce communication barriers faced by deaf and mute individuals.
YOLOv10-Based Detection: Applies a state-of-the-art object detection model to Bangladeshi Sign Language recognition.
Custom Dataset: Uses a curated dataset of 1,949 labeled images covering 14 unique static signs.
Accurate Recognition: Achieves an overall average accuracy of 90.67% across the supported signs.
Practical Application: Provides a foundation for developing accessible sign language communication tools.

Social Impact

This project aims to contribute to a more inclusive society by addressing communication barriers faced by people who rely on Bangladeshi Sign Language. By translating hand gestures into text, the system has the potential to improve everyday communication, reduce social isolation, and support greater participation in education, employment, and community activities.

The research also contributes to the development of sign language recognition technology and provides a foundation for future systems capable of supporting a wider range of gestures and real-time communication scenarios.

Future Improvements

Real-Time Recognition

Extend the system to support real-time gesture recognition through video input and live camera streams.

Expanded Sign Vocabulary

Increase the number of supported signs to improve the coverage of Bangladeshi Sign Language.

Sentence-Level Translation

Extend the system from individual static signs to continuous gesture sequences and sentence-level translation.

Mobile Accessibility

Develop a mobile-friendly application to make the recognition system more accessible for everyday use.

Project Highlights

This project demonstrates practical experience in deep learning, computer vision, object detection, custom dataset development, YOLOv10 model training, and sign language recognition. It combines technical innovation with a socially meaningful objective: improving communication accessibility for people who rely on Bangladeshi Sign Language.

GitHub Repository

The complete source code of the project is available in the GitHub repository.

GitHub Repository