Real-time Hand Gesture Recognition Using MediaPipe and Artificial Neural Networks

Hand gestures are an important form of communication, especially for individuals who use American Sign Language (ASL) to communicate. This study explored the use of Hand Gesture Recognition (HGR) using a dataset of 135,000 images, with 27 classes representing the letters A to Z and the space character. The MediaPipe framework and an Artificial Neural Network (ANN) with four hidden layers were used for building a gesture recognition system. Also to prevent overfitting, two dropout layers. In the input and hidden layers, the Rectified Linear Unit (ReLU) activation function was utilized, and the softmax function was used in the output layer to predict probabilities for each class and obtained an accuracy of 99.34%, indicating the effectiveness of this approach for HGR. This study has important implications for improving communication for individuals who use ASL and may lead to the development of more advanced gesture recognition systems which will be beneficial to those who are deaf or dumb.