Gesture Recognition based on EMG Signals: A comparative study

Abstract

The prosthetic industry of the world focuses mainly on the development of prosthetics for one arm amputees. An idea to make this prosthetic hand more intelligent is to capture any gesture apart from predefined one is converting the brain waves or the electromyographic signals to transport signals then to the hand muscles which will eliminates the mechanical use of other hand to control the movements completely and will be helpful for disabled person. In recent years, there has been a tremendous interest in introducing machine learning techniques to the growing data to gain a better result which inspire author to study and find an effective machine learning method for classifying Electromyogram (EMG) signals by applying feature extraction and classification methods. A prototype for the implementation of gesture on the prosthetic hand is also prepared using Arduino and DC motors and we have captured a real time dataset as mentioned below. Using the same, a deep comparative study of multiple classification techniques is performed to obtain the performance of different models along with its log loss. Finally, comparative results are presented using various performance measures such as sensitivity, specificity, accuracy, F-measure and area under ROC curve (AUC).