The Hand Gesture Recognition Based On The Convolutional Neural Network

In the past ten years, Internet has been rapidly developed, which brings the rapid development of the related industries. One of them is the human interaction, in which gestures recognition become a popular way for such interaction. Gestures will gradually replace the keyboard and mouse input as a new format of computer interaction in the future.Gestures also have their own natural advantages. They are not only easy to learn but also easy to use. The gesture recognition has been applied to many circumstances now, such as human-computer interaction, automatic sign language recognition and medical applications, and so on. Recognition based on the application of hand gesture can create the user interface more intuitively. The research in this area has become one of the hot topics in natural human-computer interaction.The convolutional neural network is a multi-layer perception that is used for identify two-dimensional images. Unlike the traditional hand gesture recognition that needs a complex set of feature extraction, the characteristics of the convolutional neural network can be used as an alternative to detector. It can directly accept the input of the preprocessed two-dimensional image and can effectively distinguish them.This paper presents the algorithm based on gesture recognition of convolutional neural networks by constructing a seven-layer network construction and extracting characteristics twice to reduce the false rate. By comparing this new method with traditional gesture recognition algorithms based on Hu moments and BP algorithm, this paper showed that the new gesture recognition algorithm can extract the characteristics of the graph patterns. On the other hand, our algorithm has also compared with other gesture recognition algorithms based on the traditional neural network to show that this novel algorithm can improve the recognition rate and robust of the algorithm, Through experiments, we have also shown that the influence of numbers of character pictures and the changed empirical value on the recognition rate of the neural network.