The Recognition Method Of Sequential Combination Actions Based On Surface Electromyography And Acceleration Signals

Through signals processing methods and pattern recognition algorithms,the Human Machine Interface(HMI)based on neural electric signals realized a convient,natural and robust interaction method between human and intelligent devices.In order to supersede the conventional ways,such as mouse,keyboard and joystick,these neural control systems must ensure the number of input commands and the accuracy of commands recognition.Surface electromyography signals(sEMG)directly reflect muscle condition and motion intention and acceleration signals exhibit the direction,velocity and displacement of movements.Both of them are ideal to new-type HMI.This paper employed sEMG signals and the combination of sEMG and ACC signals as the input to recognize 17 kinds of hand gestures and 3 kinds of complex actions.Including 6 kinds of wrist actions and 11 kinds of finger actions,the 17 kinds of hand gestures were recognized by a LIB-SVM classifier with more than 95% accuracy.Specifically,six-channels sEMG signals were filtered by the Chebyshev ? filter with the bandpass from 25 Hz to 450 Hz.Moreover,9 kinds of time domain features were extracted by sliding windows and PCA algorithm were employed to feature reduction.Finally,the results were validated by 10-fold cross validation with 100 times.16 kinds of arm movements were recognized with an average accuracy of 96.13%.One low-pass Chebyshev ? filter was utilized and the feature was also extracted by sliding windows in time domain.As to the 3 kinds of sequential combination actions,they were recognized with two steps.The action primitives,which made up the complex actions,were divided into hand gestures or arm movements with 100% accuracy at the first step.Then they were recognized by the classifiers which were trained by the hands gestures and arm movements samples,with an average accuracy of more than 80%.