Gesture Recognition Based on YCbCr Color Space and Neural Network

Abstract

Aiming at the problem of gesture recognition in complex background, a convolutional neural network gesture recognition algorithm based on improved YCbCr color space is proposed. Firstly, 8000 images of four gestures in the complex background are collected as the data set of this study. Then, the data set is preprocessed based on YCbCr color space, and the adaptive threshold method is used to improve it. Finally, a shallow convolutional neural network is built and trained with the preprocessed data set. The experimental results show that the gesture recognition accuracy of this method can reach 98.2% on the collected data set, which is higher than 85.7% and 90.6% using AlexNet and VGG-16.