Automated recognition of Myanmar sign language using deep learning module

The recognition of sign language in real-time poses a significant challenge due to the dynamic nature of hand gestures, and determination of the start and end points of each gesture becomes difficult in a video stream required in the recognition system. For diverse range of applications, researchers have undertaken vision-based gesture recognition studies. In this current research, deep learning approach is employed to develop a novel model specifically designed for the classification of sign language. Our study focuses on generating datasets comprising videos of Myanmar sign language (MSL) at the word level, incorporating various lighting conditions and complex backgrounds. The implementation of a Convolutional Neural Network (CNN) model allows us to present a unique approach to hand gesture recognition, using a comprehensive process involving data collection, parameter analysis, CNN model creation, as well as training and testing phases, aimed at facilitating precise recognition tasks. Notably, the utilization of the Stochastic Gradient Descent with Momentum (SGDM) optimizer resulted in a remarkable 98% success rate, while the Adam optimizer demonstrated an even higher success rate of 99%. Therefore, our proposed system is well-suited for the development of wearable real-time systems capable of automatically recognizing MSL.