Real-Time Model for Dynamic Hand Gestures Classification based on Inertial Sensor

Hand Gesture Recognition (HGR) is one of the vital topics of research in health and human-machine interaction in recent years. HGR is required for sensitive applications. Unlike static recognition, real-time gesture recognition needs the successive analysis of gestures. The design of effective gesture recognition and prediction systems using machine learning methods has attracted considerable attention. Most of the proposed solutions were not satisfied in terms of accuracy and the real-time testing process for dynamic gesture recognition. The aim of this work is to design a real-time Artificial Neural Network (ANN) model to classify dynamic gestures based on an Inertial sensor. This model is founded on a full control system with high resolution, real-time response, wireless, compact, and high sensitivity ensured with Wi-Fi communication with the ESP32 board. Based on the implemented smart system, one subject participates in collecting the dataset and performing four dynamic gestures. The developed ANN model successfully classified four dynamic gestures with an overall accuracy of 96%. Moreover, online testing and validation steps were added, and overall accuracy of 67% was achieved in 6 seconds, producing 213 real-time predictions. The overall performance of online testing has to be investigated by adding more subjects and sensors to increase the real-time prediction rate.