Hypertuned Deep Convolutional Neural Network for Sign Language Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Sign language is crucial for communication for individuals with speech and hearing impairments.
• American Sign Language (ASL) recognition presents challenges due to high intra-class similarity and complexity.
• Existing methods struggle to achieve high accuracy in ASL recognition tasks.

Purpose of the Study:

• To develop and evaluate a Deep Convolutional Neural Network (DeepCNN) for accurate ASL alphabet recognition.
• To address the challenges of intra-class similarity and complexity in ASL recognition.
• To improve the performance of ASL recognition systems through deep learning techniques.

Main Methods:

• Utilized a Deep Convolutional Neural Network (DeepCNN) architecture for ASL alphabet recognition.
• Implemented data augmentation techniques to artificially increase the size and diversity of the training dataset.
• Trained and evaluated the DeepCNN model on a standard ASL dataset.

Main Results:

• The DeepCNN model demonstrated consistent performance on the ASL dataset.
• Data augmentation significantly improved the DeepCNN model's performance.
• The proposed DeepCNN achieved substantial accuracy gains compared to state-of-the-art approaches, ranging from 3.26% to 19.84%.

Conclusions:

• Deep Convolutional Neural Networks are effective for American Sign Language alphabet recognition.
• Data augmentation is a vital technique for enhancing the performance of DeepCNN models in this domain.
• The developed DeepCNN approach offers a significant improvement over existing methods for ASL recognition.