EMG based gesture recognition using feature calibration

We propose a feature calibration method to improve the electromyography(EMG) based gesture recognition accuracy robust to noise. The feature calibration is done by subtracting the estimated noise power calculating during a resting period. For the accurate noise power estimation, we adopt the minimum statistics(MS) based noise estimation. For the performance evaluation, we compare the recognition accuracy of the feature calibration with that of no calibration. As noise level becomes higher, the proposed feature calibration shows clear improvement in accuracy. When SNR is 0dB, the proposed method shows 50% improvement in average accuracy.