A new method to estimate muscle fatigue quantitatively from surface electromyography (EMG) is proposed. The ratio of mean frequency (MNF) to average rectified value (ARV) is used as the index of muscle fatigue, and muscle fatigue is detected when MNF/ARV falls below a pre-determined or pre-calculated baseline. MNF/ARV gives larger distinction between fatigued muscle and non-fatigued muscle. Experiment results show the effectiveness of our method in estimating muscle fatigue more correctly compared to conventional methods. An early evaluation based on the initial value of MNF/ARV and the subjective time when the subjects start feeling the fatigue also indicates the possibility of calculating baseline from the initial value of MNF/ARV.

No abstract

A new method to estimate respiratory signal from thoracic impedance is proposed. To realize battery powered, wearable respiratory monitoring devices, low current impedance measurement techniques are desired. However, under low current conditions, conventional methods to separate cardiac and respiratory signals do not work well as the cardiac signal is much larger than the respiratory signal. In the proposed method, respiratory signal is estimated by calculating an envelope curve from the detected T waves of cardiac component. The results of the experiments show that the accuracy of proposed method is greater than conventional method.

In this paper, a novel method of human identification using electrocardiogram (ECG) is proposed. In the method, while normalizing RR interval, in addition to normalized signal where time interval of P wave, Q wave, R wave, S wave relatively to R wave is unaligned, normalized signal where time interval of those peaks is aligned is also generated. Wavelet transform is then applied to both normalized signals and feature vector is extracted from their wavelet coefficients. ECG data are collected from 10 subjects using a pair of dry electrodes which are held by two fingers. Experiment results show that adding wavelet of peak-aligned ECG improves the classification accuracy, where the maximum accuracy is 100%, 97%, and 90% for data measured in more than 20 seconds, 5 seconds, and 3 seconds respectively.

No abstract