Research On Human Lower Limb Motion Recognition Based On Surface Electromyography

With the increasing of global aging,the proportion of disabled people continue to increase.Lack of mobility not only increase the burden of community and family,but also make disabled people less confidence in life.To enhance the independent mobile ability of the crowd,lower limb exoskeleton robot,active orthosis,artificial neural prostheses and other rehabilitation exercise auxiliary equipment gradually grow and developed.With the development of sensor technology and motion recognition technology,the use of bioelectric signals to identify the human body motion intention to control the movement auxiliary equipment has attracted much attention.Surface electromyography(EMG)signal not only contains rich information,with a high signalnoise ratio,and is directly related to human joint movement,especially suitable for human motion identification research.At present,most research mainly concentrate in the EMG based motion pattern recognition,less for the lower limb continuous motion recognition.The main contents of this paper include the following aspects:(1)The experimental environment was built to detect four channels EMG signals and knee joint flexion/extension angle from different motion patterns(standing swing leg,sitting swing leg,walking straight line and stand to sit to stand movement).At the same time,how the EMG signal processing methods make a performance on muscle activation were discussed,the correlation coefficient between EMG signals and knee movement are also analyzed.(2)The combinations of different EMG signals were established,and the highorder polynomial model and its improved model were used to map various EMG signal combinations to the knee joint angle,then model parameters were identified by least square method,then how the combination of EMG signals,model selection or model degree,and EMG signal processing method have an effect on knee joint angle estimation of four motion patterns were discussed,at last,the best model degree and it best EMG selection were chosen.The results show that both the two high-order polynomial models have a good performance on knee joint angle estimation of walking straight line movement and sitting swing leg movement,but the improved high-order polynomial model relatively has a better estimation performance.At the same time,it may not be better for the EMG based knee joint angle estimation where EMG signal have a higher linear correlation with angle.(3)After determining the optimal EMG combinations,the neural network was used to make a further analysis of the performance for EMG based knee joint angle estimation.Three different neural input cases were made to compare the joint angle estimation performance,the distribution of the segment correlation coefficient method is prompted to evaluate the local estimation performance of neural network,and the number of hidden layer neurons were optimized.The results show that though increasing the number of hidden neurons can reduce the estimation error,but it may not improve the local performance of knee joint angle estimation.At the same time,when compared to the high-order polynomial model,which have a relative similar performance for knee joint estimation of four kinds of motion pattern,and have almost the same effect on the walking straight line movement and swing leg movement,but may not suitable for knee joint angle estimation of the other two motion pattern,but in general,the neural network has a better estimation performance than the high-order polynomial model.In addition,both the two EMG processing method seem to have no difference on knee joint angle estimation no matter using neural network model or highorder polynomial model.This study has important reference value for how to choose and assess the best EMG features,motion recognition model or algorithm selection during the development of myoelectric recognition system.