sEMG-Based Estimation of Knee Joint Angles and Motion Intention Recognition

The research of surface Electromyography (sEMG) has developed rapidly in recent years, mainly in the recognition of motion intention and the prediction of the joint angles. Consequently, sEMG has come to the foreground as an ideal human-computer interface. This paper proposes a method which applies the sliding window based on energy value of the first differential signal (DSWE) to achieve a more stable effect for determination of onset time. A combination of 4 time domain features and 19 fuzzy entropy of wavelet subspaces are selected as features for recognition. Afterwards, a classification and an estimation model were established utilizing BP neural network. To verify the effectiveness of the BP neural network, three motion patterns including standing up, flexion and extension were conducted respectively. The result presents that the average recognition accuracy of human motion intention is 98.3% and the 3-channel system can estimate knee joint angles with 3.25-11.65% error.