Research On Motion Recognition And Prediction Of Angle From SEMG And Accelerometer Signals Of Lower Limb

Among the handicapped,extremity deformity has the largest ratio.Besides,because of some reason the aged may suffer from locomotor impairments.Large amount of assistive devices are in needed.Neural control of computerized,intelligent prosthetic legs is under research.Since surface electromyographic(sEMG)signals are responsible for the muscle movements and non-invasive.It's regarded as a control source of intelligent artificial limbs or other assistive devices.This paperis to recognize the motion pattern and predict the angle trajectory based on sEMG.So that it can be applied to the control ofintelligent artificial limbs.SEMG and accelerometer signals are collected together from the lower limb.Then these signals are processed to recognize the motion pattern.And angle prediction models are established based on each detected motion to get the exact motion of lower limb.The major works completed includes the following four aspects:(1)Since the lower limb's sEMG is too complicated.SEMG and accelerometer signals fusion is adopted to get more movement information of the lower limb.The signals are collected from four normal subjects and one lower leg amputee.Curve fitting based on median filtering was applied to reduce accelerometer noise.A method to detect movement onset is used based on correlation coefficients of sEMG power spectrum.(2)Extract features from sEMG and accelerometer signals,then the kernel linear discriminationanalysisis applied to making the classification.SEMG is analyzed in terms of time domain,autoregression coefficients and the wavelet coefficients.The principal components feature based on wavelet coefficients is then used.For the accelerometer signals,features based on dynamic time warping distance isextracted.At last,use the kernel LDAto evaluate the validity of the fusion feature.The fusion-based feature outperformed the classifiers with only sEMG signals or accelerometer signals.(3)Support vector machine is used as the classifier.RBF kernel function is chosen as well.Further improvement is made to optimize the SVM parameters by improved grid search algorithm.To improve the recognition rate,the post-processing methods of majority vote is typically used.These methods c Compared to kernel LDA classifier,the performance of SVMbased on improved grid searchis better.(4)Angle prediction can provide continuous control signal,GRNN algorithm is used topredict angle trajectory of thigh corresponding to the present motion pattern.Fruit fly optimization algorithm is applied here to optimize the spread parameter of GRNN.By the fusion and analysis of sEMG and kinematics parameter,five lower limb motion patterns can be recognized accurately.Besides,angle prediction model of each movement is established,which is helpful to the control of EMG-prosthesis in the future work.