KuSL2023: A standard for Kurdish sign language detection and classification using hand tracking and machine learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Sign Language Recognition (SLR) is crucial for deaf and hearing-impaired communication.
• A significant gap exists in resources for Kurdish Sign Language (KuSL).
• Existing datasets lack standardization for KuSL detection and classification.

Purpose of the Study:

• To establish a comprehensive standard for KuSL detection and classification.
• To create a real-time KuSL recognition dataset.
• To evaluate machine learning models for KuSL recognition.

Main Methods:

• A new KuSL dataset was created by merging and refining ASL and ArSL2018 datasets, totaling 71,400 images.
• The dataset includes 34 Kurdish sign categories with diverse lighting, angles, and backgrounds.
• Machine learning models including CNN, KNN, and LightGBM were used for performance evaluation.

Main Results:

• The Convolutional Neural Network (CNN) model achieved a high accuracy of 98.22%.
• Traditional classifiers like KNN (95.98%) and LightGBM (96.94%) demonstrated competitive accuracy with faster training times.
• The developed KuSL dataset proves robust, efficient, and accurate.

Conclusions:

• The new KuSL dataset sets a benchmark for Kurdish Sign Language recognition.
• High accuracy and efficiency achieved pave the way for real-time applications.
• This work supports the development of assistive technologies for the deaf community and advances gesture recognition.