Research Of Hand Gesture Recognition Based On Convolutional Neural Network

Gesture is not only a natural and efficient way of interaction,but also the only way to interact with normal people for deaf-mute.However,in most cases,ordinary people cannot communicate with them by sign language,thus it is necessary need for intermediaries to translate.If there is a software that can help ordinary people easily get the meaning of sign language by translating their sign language into words,it will greatly improve the interact efficiency of the deaf-mute and the outside world.In this paper,the CNN algorithm is used to realize gesture recognition.Development process and research progress of gesture recognition are first analyzed,and by comparing with other depth learning algorithms,convolution neural network is chosen as the basis of gesture recognition algorithm.Working theory and training process of CNN are analyzed in detail based on the latest research results,and factors which can influence the system performance are clarified,such as network structure,convolution kernel size,pooling method,pooling size,activation function and learning rate and so on.Then we take measures to promote system performance based on the LeNet-5 model,and then factors may affect the capability of the neural network were analyzed by quantitative analysis on MNIST and CIFAR-10.The experimental scheme is as follows: data set is preprocessed before training,and then the CNN is trained by the gray image with less information and compared with therecognition result of data set without preprocessing.The experimental results show that the data set after preprocessing has the advantage in recognition accuracy and recognition speed,which verifies the validity of the experimental scheme.The final mean recognition accuracy rate is 93.8%,which has practical application significance and can provide reference for the future deaf-mute communication system.