A Static Sign Language Recognition Method Enhanced with Self-Attention Mechanisms.

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Wearable devices face challenges in static sign language recognition accuracy, noise resistance, and robustness due to user diversity.
• Existing methods struggle with variations in gesture execution and environmental noise.

Purpose of the Study:

• To develop a novel static sign language recognition method that enhances accuracy and robustness for diverse users.
• To improve the performance of sign language recognition systems in real-world, variable conditions.

Main Methods:

• A self-attention mechanism is employed to highlight key features for gesture classification.
• A convolutional neural network (CNN) is utilized for feature extraction and classification.
• The method integrates a weight function to focus on critical sign language features.

Main Results:

• Achieved 99.52% average accuracy on a standard 36-gesture American Sign Language dataset.
• Demonstrated robust performance under angular bias conditions (98.63% at ±9°, 86.33% at ±18°).
• Outperformed existing methods in noise resistance and robustness for static sign language recognition.

Conclusions:

• The proposed self-attention enhanced method significantly improves static sign language recognition.
• The system exhibits superior noise resistance and robustness, making it suitable for diverse user populations.
• This approach offers a promising solution for more accessible and reliable sign language communication technology.