Video Sign Language Recognition Based On Neural Network

China is a country with a large number of hearing impaired people.Sign language is the only language for deaf-mutes to communicate with the outside world.With the development of society,people with hearing impairment want t o make use of the power of science and technology to achieve zero distance c ommunication with the outside world.In recent years,the rise of the artificial intelligence industry has made this aspiration possible.The practical environme nt of sign language is complex and changeable.And this have made the study of sign language difficult.Although the traditional sign language recognition h as also made great progress,it requires the sign language users to wear some proprietarydevices in the process of using them.These devices are usually very expensive.This may limit the generalization of the traditional method of sign language recognition.In the research of static sign language recognition,this paper combines the machine learning method with the traditional Hu invariant moment method,and proposes a static sign language recognition model(kNN-Hu).The method is applied to static sign language recognition,and the experimental results show that the recognition rate of this method reaches to nearly 98%.This thesis focuses on the dynamic sign language recognition.The thesis c ombines the network model of deep convolution neural network with long-short term memory recirculationneural network to recognize the video sign language.And on this basis,this thesis makes an improved network model to get a vid eo sign language recognition model that can be used.The biggest advantage of this model is that it can directly use the data obtained by ordinary mobile ph one camera as an input.Sign language users do not need to wear data gloves or colorized gloves.We do not need to preprocess the inputted video.This pro vides the conditions for the promotion of this method.The main work of this thesis:1.On the basis of the combination of the kNN method in machine learning with the traditional Hu invariant moment,a static sign language recognition method is proposed,and its feasibility is verified experimentally.2.We shoot a set of sign language video datasets consisting of 11,172 videos.The length of each video ranged from 1 to 3 seconds,and each video contains a complete word action.The video datasets are collected by 7 people independently.3.Based on the network model combining 3D convolutional neural network and LSTM network,an improved network model is proposed and tested.Experiments show that the network model can improve the recognition accuracy.