การรู้จำตัวอักษรภาษามือแบบอเมริกันโดยใช้เทคนิคการผสมเชิงสถิตและพลวัตของการจำแนกร่วมกันร่วมกับโครงข่ายประสาทเทียมแบบสังวัตนาการ

Abstract

This thesis proposes the American Sign Language Alphabet Recognition Using a Static and Dynamic Mixing Technique of Ensemble Classification Combining with Convolutional Neural Networks to recognize the American Sign Language alphabets that have both dynamic and static hand gestures. Sign Language is an important tool to communicate between people who impaired hearing. So, this model is evaluated to help and interpret the gesture to character or text that normal people is able to understand. Then Generative Adversarial Networks (GANs) method is used to generate the synthetic data to increase the diversity and size of dataset. In this ensemble approach, consists of Convolutional Neural Networks that combined with 3 methods are 1. Support Vector Machine (SVM) 2. k-Nearest Neighbor (k-NN) and 3. Long Short-Term Memory (LSTM), then using the voting method to classify the final output. The experimental results demonstrated that the accuracy of the proposes method is over 99% and also this model can recognize both static and dynamic hand gestures that cover all of American Sign Language alphabet gestures.