Description

Human gestures are immensely significant in human-machine interactions. Complex hand gesture input and noise caused by the external environment must be addressed in order to improve the accuracy of hand gesture recognition algorithms. This nature requires a different approach from action recognition to capturing the complex spatial features based on the Local Binary patterns algorithm. Many gesture categories have very similar motion or temporal patterns posing a challenge on temporal processing. To overcome this challenge, we employ a combination of 2D-FFT and convolutional neural networks (CNN) in this process.