Abstract:

Electromyogram (EMG) signals are signals that contain information about contractions in the muscles. EMG signals are personal and express which muscles contract at what intensity. In detecting these signals, Myo armband has been used frequently in recent years. There are eight EMG sensors, accelerometer sensors and gyroscopes on the Myo armband. These eight EMG sensors settle on different muscles on the arm and measure the contraction intension of the muscles during gesture. In this way, the gesture using the information of which of the eight sensors is contracted can be recognized. Myo armband acquire EMG data with a sampling frequency of 200 Hz. In this study, EMG data was acquired by repeating 10 times 4 different hand gestures by 4 subject by attaching Myo armband to the right forearm. First, a high pass filter was applied to eliminate the noise from the acquired data and then the times when the hand gesture started and ended were determined. The aim of this study is to propose a new method to the literature to find the start and the end times of hand gesture at this point. Five time domain features of the preprocessed EMG signals were extracted. These features were root mean squire (RMS), mean absolute value (MAV), zero crossing (ZC), waveform length (WL) and slope sign change (SSC). Sequential forward selection was made in order to find the most successful feature set among the extracted features. For classification, SVM and KNN algorithms were used. As a result of the study, SVM algorithm with the WL feature gave the best result and 98.75% performance was achieved. The result obtained was compared with the studies in the literature. In addition, other methods in the literature used to find the times when the gesture starts and ends were applied to the dataset used in this study and the results were shown.

Keywords: