Hand movements classification based on Myo armband signals

Many myoelectric pattern recognition algorithms have been applied to predict human movement intent. Recognition of human motion intent is used to control the prosthetic hand. However, this field is still far from achieving the needs of real life. In this study, 17 hand movements were classified based on Myo armband signals. Myo signals are obtained from database-5 (DB5) of the NinaPro dataset, a publically available dataset for hand prosthetics. In this study, a method for analyzing Myo signals of DB5 is proposed to perform short-term hand gesture recognition. The Myo signals of DB5 are long-term signals that include multiple gestures with their repetitions. The proposed approach focuses on segmenting the long-term signals of DB5 into several short-term signals. The short-term signal represents a muscle activity of a single gesture. Five statistical time domain features are extracted from the short-term signals. The results achieved a high classification performance of about 86.5%, 83%, and 82.2 % as an average precision, sensitivity, and F1-score, respectively, for 17 gestures. The classification accuracy is examined for a different number of gestures based on the support vector machine (SVM) classifier. A comparison with existing literature works shows that short-term identification ensures high classification accuracy than long-term identification.