Sign Language Recognition for Deaf People using Novel CNN Compared with GCN and BERT
Geetha Reddy, S. John Justin Thangaraj, Sai Kanna R, A. Rajalingam
Abstract
The aim of the proposed study is to develop a Sign Language Recognition system for deaf individuals by employing CNN and enhancing accuracy through the integration of Graph Convolutional Networks (GCNs) and Bidirectional Encoder Representations from Transformers (BERT). The CNN model is used to identify letters in sign language. The study compares the performance of the CNN model with a combined GCN+BERT model in terms of recognition accuracy. The CNN model was developed and evaluated with a sample size of 325 per group, determined using a power analysis with a value of 0.8. The results show that the CNN model achieved a higher accuracy of 92% with a lower mean error compared to the GCN+BERT model, which achieved 85% accuracy. The findings indicate that CNN provides superior accuracy in sign language recognition compared to the GCN+BERT model.