Signed language

Hand Gesture Recognition (HGR) Overview

• HGR enhances human-computer interaction (HCI).

• Importance of accurate character recognition for effective communication.

Deep Learning Approach

• Utilization of Convolutional Neural Networks (CNNs): Modified AlexNet and VGG16.

• Focus on recognizing American Sign Language (ASL) characters (both alphabets and numerals).

• Features extracted via pre-trained CNN models followed by classification using Support Vector Machine (SVM).

Performance Evaluation

• Achieved recognition accuracy: 99.82%, surpassing many state-of-the-art methods.

• Two validation methods used:

  • Leave-one-subject-out.

  • Random 70-30 split.

Gesture Recognition Challenges

• High inter-class similarity poses challenges for recognition accuracy.

• Some characters often misclassified due to similar gestures.

Experimental Analysis

• Dataset: 36 ASL characters with a total of 2520 images from 5 subjects.

• Resizing of images for compatibility with CNN inputs.

• Data augmentation applied to expand training dataset.

System Architecture

• SVM used for classification, leveraging limited memory by focusing on support vectors.

• Highlighted need for robust recognition strategies across similar gestures.

Future Directions

• Explore attention-based CNN architectures to improve differentiation among similar gestures.