The Recombination of Gesture Nodes Enables Higher Accuracy in Small Data Sets

Abstract

The hand gesture conveys many information such as hand sign language, detailed operation by hand, and gesture command in policy and army. In VR application, hand motion capture becomes more and more important in order to build the immerse virtual experience. The existing digital hand glove mainly works on the key point tracking of hands in real time. However, the immerse VR system needs not only the changes of the hand position movement but also the meanings of its movement and gesture in order to create the realistic interaction. In this paper, we treat this issue as a classification problem in supervised learning. We use Principal Component Analysis (PCA) to reduce the dimension of original input and analyze its result. We recombining the new data and raw data according to analysis, reduce not only the amount of computation but also the dataset. In the experiment, we implement our method on Raspberry Pi 3 Model B+. The results demonstrate that our method successfully classify 12 different hand gestures in the accuracy rate is over 93.5% less than 1ms in the immerse system and with just 500 training data.