Static Sign Language Recognition System Based On Convolutional Neural Network

Sign language is a tool that deaf and dumb people use to communicate with the world.For deaf and dumb people,some of them can't hear the beautiful voices of the world,and some can't speak beautiful words to others.The main way they express their inner thoughts is sign language vocabulary that has various meanings by the combination of finger and limb movements,these sign languages are filled with their true feelings.But in real life,the number of ordinary people who are proficient in sign language is not large.For most people,there is a huge obstacle to communicating with deaf and dumb people.They cannot understand the meaning of sign language of deaf and dumb people.Therefore,the recognition of the sign language of deaf and dumb people is of great significance for deaf and dumb people to communicate with the world.Aiming at this problem,this paper designs and sets up a static sign language recognition system based on artificial intelligence deep learning technology.Sign language recognition is divided into two parts,hand feature extraction and hand feature model training.For hand feature extraction,starting from the traditional hand extraction algorithm,this paper analyzes and verifies the hand extraction algorithm based on Gaussian filtering,the hand extraction algorithm based on the RGB color space,and the hand extraction algorithm based on the YCrCb color space.Aiming at the disadvantages of several commonly used algorithms,a new hand feature extraction method based on improved watershed algorithm was proposed and verified.This method has achieved good experimental results in experimental tests.Aiming at the hand feature model training part,this article starts from the theory of artificial intelligence machine learning,analyzes and verifies the calculation process of neural networks in the field of deep learning in detail,and deals with the main operations in convolutional neural networks,such as convolution,pooling,and batchnormalization for in-depth analysis and formula derivation.In order to achieve the goal of hand feature model training,a complete sign language feature extraction network is designed and built.In view of the problem that the deeper the network exists in the neural network,the more difficult the feature training is,the residual network structure is introduced into the neural network.The neural network can extract more information from the sign language data set,and help the trained model to quickly and accurately extract the actual sign language information.When designing and building the sign language recognition system,the two major parts of sign language recognition are combined.Use the camera as the input of the recognition system to extract hand information for each frame of the video input by the camera,and then compare the extracted information with the sign language information stored in the successfully trained sign language model to obtain the sign language input message by the camera,and output the message for users' reference.The sign language recognition system designed and constructed in this article has been verified in practical applications,and can recognize and display the sign language in the camera screen in real time,which is highly practical.