Feature based weighted neural network for hand gesture recognition

Abstract

Recently, developing interaction techniques that allow gesture recognition for home application and game control is one popular field in Human-Computer-Interaction (HCI) research. In this paper, we proposed a method for hand gesture recognition using artificial neural network algorithm commonly used in HCI. Previous studies have generally used distinguished distribution datasets. However, in the real world, gesture data are imbalanced. In addition, if gesture distribution data are imbalanced, it is difficult to classify gesture. To solve these problems, we present a gesture distribution feature based weighted Neural Network (FWNN) after adding trajectory distribution data to input layer before network training. Specific hand trajectory coordinate and the number of extrema of hand trajectory are added to the trajectory distribution data. Our experimental results demonstrate that our method is much more accurate than the method of using only sequence data.