Research On Continuous Motion Estimation Of Multiple Joints Of Upper Limb Based On SEMG

Surface electromyography(s EMG)is a kind of bioelectrical signal,which is formed by superimposing the potential difference from the muscular contraction and relaxation on the skin surface.It is closely related to the activity of muscles and can be used to reflect the movement of human body.As s EMG signal contains abundant information and is easy to collect,the research of human motion estimation based on s EMG signal has been successfully applied in the fields of exoskeleton robot,prosthetic control,human-computer interaction,medical rehabilitation and so on.The continuous motion estimation of human upper limb is to estimate the motion angle of each joint of upper limb by analyzing and calculating the EMG signal.In this paper,the wrist joint,elbow joint and finger joint of human upper limb are taken as the research object,combined with the muscle synergy theory,and finally the gnmf-svr model is constructed to map the human EMG signal to the joint angle.The specific research work is as follows:(1)This paper designed an experiment of continuous motion estimation with multiple joints.The Trigno wireless EMG acquisition system was used to collect the surface EMG signals of the human upper limb wrist,elbow and finger joints,and the Codamotion 3D motion capture system was used to collect the angle signals of the three joints.In order to obtain a more accurate estimation of the joint angle,the collected original EMG signal and joint angle signal are preprocessed using high-pass filtering,full-wave rectification,and low-pass filtering.(2)The theory of muscle coordination is quantified by graph regularized nonnegative matrix decomposition,and the cooperative elements of independent actions are extracted from EMG signals.The differences between graph regularized nonnegative matrix decomposition and traditional non negative matrix decomposition in extracting synergetic elements are compared through experiments.(3)Based on the calculated synergy element,the non-negative least square method is used to obtain the muscle activation coefficient.On the premise that the input is the muscle activation coefficient,the accuracy of joint angle estimation of LSTM,BP neural network and support vector regression is compared.And the influence of channel missing on the overall performance of the model is explored.The experimental results show that support vector regression has a better estimation effect on the test data.However,in terms of stability at different speeds,the LSTM performs better.The absence of biceps brachii and flexor carpi channel signals has a great influence on the accuracy of predicting the corresponding movements.