Dynamic Hand Gesture Recognition Based on 3D Convolutional Neural Network Models

Abstract

Hand gesture is a natural communication method which could be used to create a more convenient interface for human-robot interaction. In this study, we use the simplest laptop camera as an input sensor. We designed a 3D hand gesture recognition model. The model is trained with the Jester dataset. After being trained about one day in a MacBook Pro (i5 2.3GHz), the model reached an average accuracy of 90%. We built a web application that implements the hand gesture recognition system and provides the recognition service to users.