Tools
Tools
Bahadkar, Siddhant Sanjay (2025) A Deep Learning Approach for American Sign Language Alphabet Recognition. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (4MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
Communicating with individuals who are deaf or have speech impairments remains difficult when sign language cannot be assumed to be widely known. This project proposes a fully automated, real-time American Sign Language (ASL) recognition pipeline, leveraging deep learning and computer vision. Our objective was to identify and benchmark several advanced convolutional neural network (CNN) architectures for distinguishing the full ASL alphabet, and then to embed the most accurate model into a live recognition stream driven by a webcam and MediaPipe hand tracking. This study trained and fine-tuned a suite of pre-trained models like VGG16, ResNet50, InceptionV3, DenseNet201, and MobileNetV2 alongside a custom CNN on a balanced ASL dataset of 27 handshape classes. Each architecture was rigorously quantified with classification metrics and real-time latency. VGG16 emerged as the top performer, achieving an overall validation accuracy of ≈93.2%, with stable inference speeds during live demonstrations. Visual diagnostics and confusion matrices revealed both the discriminative strengths and shortcomings of the various designs. These findings affirm that transfer learning, when coupled with fine-tuned parameter adaptation, yields proficient ASL handshape recognisers. Embedded into a low-latency detection framework, such classifiers have the potential to broaden the reach of assistive technologies for deaf and hard-of-hearing users. Future works will transition to continuous sign tracking, gesture sequences, and multilingual lexicon support.
Actions (login required)
 |
View Item |