Motion Prediction Of Human Lower Limbs Based On EMG Signal

With the increasing aging problem in China,the population of hemiplegic patients is increasing.The traditional rehabilitation method,in which the rehabilitation physician drives the patient's damaged limb to perform exercises to help the patient's damaged limb muscles gradually regain their muscle strength,is less efficient and more physically demanding for the rehabilitation physician.To solve this problem,research has been conducted on rehabilitation assistance machines such as lower limb exoskeleton robots and active orthoses.Patients control the exoskeleton according to their own motor intention,which can better motivate patients to train and improve the efficiency of rehabilitation.The surface EMG signal is formed by superimposing the electrical signals generated by neuromuscles on the skin surface when the human body is moving.The surface EMG signal is generated about 100 ms earlier than the human muscle contraction,which is directly related to the human muscle force generation and can reflect the human movement intention in advance,and is often used as the human-computer interaction interface of intelligent exoskeletons.At present,the application of EMG signals is mainly focused on motion pattern recognition,and less applied to the estimation of continuous motion of lower limb joints.The main focus of this paper is on continuous motion estimation of lower limb joints,and the research includes the following aspects.(1)The locations of lower limb muscle groups and their roles in joint motion are analyzed,and muscle groups that are easy to acquire and have a high correlation with joint angle changes are selected.The process of surface EMG signal generation and the biological relationship between surface EMG signal and muscle force were analyzed.(2)Firstly,the surface EMG signals of the legs of the subjects were acquired during the exercise process with the correlation of joint angle changes,then the obtained surface EMG signals were implemented with noise reduction and filtering.(3)Establishing the relationship between the variation of surface EMG signal and joint angle by GA-BP neural network and extreme learning machine.(4)The accuracy of the model was further improved by limiting the filtering of the output results according to the characteristics of human lower limb movements,and the accuracy of the model was analyzed when a certain muscle of the patient could not produce the surface EMG signal,which proved that the limit learning machine model has better robustness in predicting the joint angle changes by the surface EMG signal.In this paper,we systematically study the continuous motion estimation of lower limb joint angles based on surface EMG signals,and complete the acquisition,filtering processing,feature extraction fabrication and related joint angle estimation of leg surface EMG signals.Through the result analysis the extreme learning machine can achieve the prediction of lower limb joint angle well,which has some application value for the intelligent control of lower limb exoskeleton.