Study On Prediction Of Joint Angle Of Lower Limb In Human Obstacle-Crossing Gait Based On Surface EMG Signals

Lower limb exoskeleton robot plays an increasingly important role in the fields of people’s daily assistance,military and medical rehabilitation.At present,many exoskeleton robots have some problems,such as rigid movement,too mechanical gait,too many restrictions on human joints and so on,which will lead to the wearer’s discomfort and even physical injury.Surface electromyography(sEMG)is a kind of physiological electrical signal detected from the surface of human skin during muscle activity,which can more quickly and accurately reflect the movement intention of human body.The research of exoskeleton robot based on sEMG signal technology has become a hot spot at home and abroad.In a variety of human daily gait,the sEMG signal when crossing obstacles presents strong non-linearity and instability,which makes the joint prediction of obstacle crossing gait more difficult than obstacle free gait.In order to solve these problems,this paper studies the joint angle prediction of lower limbs based on surface electromyography.First of all,through the biological theory and experiment,the muscles of lower limbs participating in the movement in obstacle free walking and obstacle crossing gait are analyzed,and the sEMG signal strength of the related muscles in the two gait is tested through the experiment,and the characteristic muscles of the two gait are selected.According to the number of characteristic muscles,sensors are selected to build the acquisition system of sEMG signal and joint angle signal.Then,the sEMG signals of each muscle in different degrees of speed were tested,and the influence of lower limb movement speed on sEMG signals was analyzed;the sEMG signals of each muscle in different obstacle height were tested,and the influence of lower limb force on sEMG signals was analyzed.Then,aiming at the problem of high frequency and irregular fluctuation of sEMG signal of human obstacle crossing gait,the method of preprocessing sEMG signal by wavelet transform and then extracting eigenvalues is proposed,and the eigenvalues of various obstacle crossing gait with different heights are extracted.Finally,BP and RBF neural network prediction models are constructed,and the feature values extracted by "without wavelet transform" and "after wavelet transform" are used as input values to predict the two joint angles of lower limbs.The prediction results show that: in terms of prediction time,BP neural network takes less time than RBF network,and has more advantages in real-time;in terms of accuracy,there is no obvious difference between the two neural networks in obstacle free gait,but in obstacle free gait,the prediction accuracy of RBF neural network is higher than that of BP neural network,and more stable.The feature value extracted from sEMG signal after wavelet transform has obvious optimization effect on RBF neural network,which is more significant in obstacle crossing gait condition,and the highest accuracy reaches 96% for knee joint and 98% for ankle joint.The prediction results of RBF neural network using wavelet transform to extract eigenvalues can meet the needs of human daily gait.