Six Sensors Bracelet for Force Myography based American Sign Language Recognition

In recent years, force myography (FMG) is considered a non-invasive method used for the recognition of different gestures such as the American sign language (ASL) by measuring presser changes due to different variations in muscle volume while performing different signs. In this work, only six FSR commercial sensors were inserted to the inner side of a bracelet to be in direct contact with the skin was used to collect the raw FMG signal from 10 healthy subjects with 10 trails. In this paper, an extreme learning machine (ELM) with cross-validation ( $\text{Kfold}=5$ ) was applied to test the accuracy of using raw FMG signal in comparison with only six time domain extracted features. The results show that the accuracy based on extracting six features was equal to 91.11%, which outperforms the raw FMG signal for gesture recognition where it reached only a testing accuracy of 85.56%.