Multi-channel sEMG Time Series Analysis Based Human Motion Recognition Method

Abstract

Towards human motion intention recognition during active rehabilitation, a multi-channel suface electromyogram(sEMG) time series based human motion pattern recognition method is proposed. An evaluation method for sEMG signal waveform similarity is designed to select the features, which are coded by a 2D Kohonen self-organization mapping net(SOM) net to get feature series. At last, hidden Markov models(HMM) are built to describe the multi-channel sEMG time series features during each motion process, and then get recognition results based on maximum likelihood estimation method for multi-model synthesis decision. This method showed a good performance on real time and accuracy in the experiment: the treadmill, elliptical and walk training modes are identified by an accuracy of 91.67 %, while the classical linear and nonlinear methods showed accuracies of 72.5 % and 88.33 %.