LADS-NET: A Deep Learning Model for Dynamic Sign Language Recognition

Deaf-mutes cannot communicate with the outside world normally due to their own reasons, while the sign language is an important medium for them to participate in social life. Intelligent sign language recognition (SLR) can not only solve the problem of communication between deaf-mutes and healthy people but also contribute to the development of human-computer interaction. The SLR method based on data gloves is not limited by the experimental site, which can fully capture various hand movements of users and extract action information more accurately. Therefore, a dynamic SLR model was constructed by integrating the Long Short-Term Memory network based on Attention Mechanism (LSTM-Attention) and the 2D Dense Convolution Network (2D-Densenet), which uses a data glove as an action information acquisition device. The spatiotemporal feature data of finger posture and palm position were collected by the data glove, and then both sets of data were fed into the LSTM-Attention and 2D-DenseNet networks for training respectively. Finally, the two kinds of features were fused by series fusion for classification and recognition, and a recognition rate of 99.13% was obtained. The experiment results show that the LADS-NET, proposed in this paper, has a higher recognition rate and better results compared with other models.