sEMG Based Movement Quantitative Estimation of Joins Using SVM Method

The sEMG based movement recognition developed rapidly in recent years, which focus on intention estimation that velocity and angle of movement joint are not concerned. This paper proposed a quantitative analysis method of sEMG, with ability to estimate motion of human joints, which can be used to control rehabilitation robot system control by patient’s own intention. The quantitative model of the relationship between sEMG signals and movement joint was established utilizing error Back Propagation artificial Neural Network and support vector machine with a Gaussian kernel, where the features of sEMG were taken as input. Considering of the actual demands of rehabilitation, the 1-DOF, 2-DOFs and 3-DOFs movement experiments were supposed to collect the information of joint angle and sEMG signals for model training. The result shows the angle prediction curve outputted by model of SVM has more than 90% consistency with the actual movement, while the model of BPNN gets a more imprecise output with complexity of movement arising. Initial online experiments on rehabilitation robot controlled by a healthy subject demonstrate that sEMG based movement control using the proposed method is feasible