1 / 2020-09-22 13:25:54

Recognition of 18 kinds of sign language based on Mechanomyography collected by 3 different devices

Mechanomyogram; Back Propagation Neural Network; Sign language recognition; Bracelet device

Sign language is the main approach for people with hearing impairment to communicate with ordinary people. To recognize the sign language motions, in this study, mechanomyogram (MMG) signals on forearm muscles were captured by three different devices and fed into a Back Propagation Neural Network (BPNN) model. The results indicated a high identification accuracy (99.0%) can be obtained for 18 kinds of sign language motions if the sensors were in stable status (embedded in a bracelet device) to collect data. The BPNN could automatically calculate the feature sets in hidden layers, which means the feature extracting step could be skipped, therefore, the recombined channels were sever as training and test set rather than feature vectors. This study could provide potential reference to the field of sign language identification, and extend to other fields like robot control, rehabilitation training, etc.