Real-Time Replication of Arm Movements Using Surface EMG Signals

Abstract

In this paper, a real time application to replicate nine arm movements is proposed. The two important joints that are controlled are wrist and elbow. Electromyogram signals are recorded for four wrist positions and five elbow positions. These signals are enhanced and features pertaining to muscle movements are extracted. Dimension of these feature sets is reduced to obtain the optimal set of features. These feature sets are given as input to the classifier. Performance evaluation of Support Vector Machine (SVM), K-Nearest Neighbors, Random Forest and Relevant Vector Machine (RVM) classifiers, in recognizing different wrist and elbow positions, is discussed. As per the results, the best overall accuracy of 93.3% was obtained from SVM with radial basis function (RBF) kernel, in classifying both the wrist and elbow positions. Although, RVM as a classifier yielded the same accuracy in recognizing wrist positions, it resulted in the lowest accuracy of 88.67% in recognizing elbow positions. Therefore, SVM-RBF fared better in identifying the arm movements. Furthermore, these arm movements are used to control the actuators.