Research On Sign Language Recognition Method Based On Deep Learning Algorithms

As an important communication tool for deaf and mute,sign language has a wide range of use value in the deaf.And the research on its complex and varied gesture can also promote the development of gesture-based human-computer interaction technology.But because of its complexity and variety of sign language,the research of sign language recognition has been very difficult.Traditional sign language recognition research methods often require signers to wear expensive data gloves for sign language data capture or to wear colored gloves which can facilitate the feature extract of sign language.Although the above methods can achieve a high accuracy under limited use conditions,but these methods are less general.It often requires the re-extract the feature manually after replacing a sign language dataset.In this paper,a series of deep learning algorithms are introduced into the sign language recognition methods.In the aspect of static sign language recognition,we propose two static sign language recognition models(SLR-CNN1 and SLR-CNN2)based on deep convolutional neural network.We use SLR-CNN1 verified the feasibility of deep convolutional neural network in sign language recognition.And the SLR-CNN2 model is used to further improve the accuracy of static sign language recognition.We introduced the global mean pool into the sign language recognition model,which greatly reduces the number of parameters and prevents the occurrence of over-fitting.Through a large number of experiments we have verified that the deep convolution neural network can automatically learn the useful sign language features,and the deep convolutional neural network can learn the subtle transformation of sign language,which can effectively identify the sign language.Finally,we also used the deep learning Caffe framework to train two deep learning sign language recognition models that could be used for actual deployment.In the dynamic sign language recognition,we combine the deep convolution neural network with the long short term memory gated recurrent neural network,and propose two dynamic sign language recognition models(SLR-LSRCN1 and SLRLSRCN2).And we modified the source code of the deep learning framework Caffe so that it can accept successive video frames as input to the deep learning model.Through a large number of experiments we have come up with a combination of convolutional neural network and recurrent neural network,which can effectively identify the dynamic sign language.Finally,in order to verify the feasibility of the deep learning algorithm in sign language recognition,we mark and labeled a large number of sample libraries that can be used for static sign language recognition by combining existing database and selfrecording database.We can more easily carry out the algorithm validation and experiment by using the above database.We have introduced a deep learning method into the sign language recognition task,which adding a strong scalability and robustness method for sign language recognition task.