Estimation of Lower Limb Periodic Motions from sEMG Using Least Squares Support Vector Regression

In this paper, a new technique for predicting human lower limb periodic motions from multi-channel surface ElectroMyoGram (sEMG) was proposed on the basis of least-squares support vector regression (LS-SVR). The sEMG signals were sampled from seven human lower limb muscles. Two channels sEMG were selected and mapped to muscle activation levels for angles estimation based on cross-correlation analysis. To deal with the time delay introduced by low-pass filtering of raw sEMG, a \(k\)-order dynamic model was derived to represent the dynamic relationship between the joint angles and muscle activation levels. The dynamic model was built by data driven LS-SVR with radial basis function kernel. The inputs of the LS-SVR are muscle activation levels, and the outputs are joint angles of the hip and knee. In experiments, 48 sEMG-angle datasets sampled from six healthy people were utilized to verify the effectiveness of the proposed method. Result shows that the human lower limb joint angles can be well estimated in different motion conditions.