Gesture Interpretation Control System Using Convolutional Neural Networks

Abstract

This paper proposes a non-invasive control system for electrical wheelchairs utilizing facial gestures of individuals captured by real-time monocular camera. The images are interpreted with a convolutional neural network that achieves up to ~99.5% overall accuracy. The control system uses an embedded system with a graphics processing unit for predicting real-time throughput with fast inference time. This solution offers great versatility, where the user can make a gesture to depict a command of his choice.