Thai Sign Language Recognition: an Application of Deep Neural Network
Anusorn Chaikaew, Kritsana Somkuan, Thidalak Yuyen
Abstract
Thai Sign Language (TSL) is the national sign language for Thai deaf people or hearing impaired in Thailand. These people with disabilities can use sign language to communicate with people with disabilities but face obstacles in communicating daily with ordinary people. Technology should play a key role in helping disadvantaged people achieve a better quality of life. This research aims to find ways to create Thai sign language recognition applications and be developed for real-time sign language translation in the next step. We purpose a simple approach with a MediaPipe framework that helps to extract the hand landmark from video on the preprocessing step and use that landmark to build the model for recognition hand gestures with various Recurrent neural networks (RNN). The result showed that the model builds with LSTM, BLSTM and GRU has an accuracy greater than 90 percent. This approach can produce an accurate close to the traditional approach.