A pattern recognition system for myoelectric based prosthesis hand control

Early myoelectric control research approaches focused on one or two degrees of freedom (DOFs). Pattern recognition has the potential to work on multiple DOFs. This paper proposes a pattern recognition method for real-time myoelectric control system. The work presented consists of EMG data acquisition, motion activity detection, data segmentation, feature extraction, dimensionality reduction, classification, and postprocessing. Support Vector Machine (SVM) and Liner Discriminant Analysis (LDA) classifiers along with five myoelectric signal features are examined and compared for constructing a feasible real-time control system. Offline and real-time testing were conducted in two separate experiments involved both body-abled and disable subjects. The SVM classifier obtained better performance with single feature sets whereas the LDA classifier achieved slightly higher accuracy for the combined multiple features. The experiment and testing results showed that the proposed pattern recognition method and the EMG data acquisition system exhibited encouraging result.