A novel model for expanding horizons in sign Language recognition

Area of Science:

• Computer Vision
• Deep Learning
• Human-Computer Interaction

Background:

• American Sign Language (ASL) recognition is crucial for communication accessibility.
• Existing computer vision and deep learning methods for ASL recognition require further accuracy and robustness improvements.
• The American Sign Language Recognition Dataset provides a large-scale resource for ASL gesture and letter classification.

Purpose of the Study:

• To comprehensively evaluate existing ASL recognition methods.
• To introduce and validate a novel deep learning architecture, Sign Nevestro Densenet Attention (SNDA), for ASL recognition.
• To enhance the accuracy and robustness of ASL gesture classification.

Main Methods:

• Evaluation of various ASL recognition techniques on the American Sign Language Recognition Dataset.
• Development and implementation of the Sign Nevestro Densenet Attention (SNDA) architecture.
• Utilizing the Nadam optimizer for efficient model training and faster convergence.

Main Results:

• SNDA achieved state-of-the-art performance with 99.76% accuracy.
• The model demonstrated perfect sensitivity and high specificity and precision.
• The fused attention mechanism in SNDA proved effective for domain-specific deep learning enhancement.

Conclusions:

• SNDA is a highly effective architecture for ASL gesture recognition.
• The developed model has significant potential to improve communication accessibility for the deaf and hard-of-hearing communities.
• Deep learning models can be significantly enhanced for specialized applications through techniques like fused attention mechanisms.