The Impact Research And Application Of EMG Gait Recognition For Mult Human-Machine Variation Scene

Accurate gait recognition and gait planning are the premise and basis for realizing the function of lower limb exoskeleton and prosthesis.With the in-depth study,lower limb exoskeleton is developing towards more intelligent and man-machine cooperation.By detecting the wearer's motional intention and improving the human-machine coordination to improve the effect of walking assistance and rehabilitation,which need effective recognition of the user's gait state during the use of lower limb exoskeletons.And adjust the control signal input of lower limb exoskeleton according to real-time gait information to realize man-machine cooperation.So,accurate gait phase recognition is very important,which is prerequisite conditions and important guarantees for manmachine motion synergy in lower exoskeleton.Existing gait research focuses on algorithms and sensors.As far as electromyographic gait recognition is concerned,the main ideas of current research are as follows: data acquisition,effective data preprocessing,reasonable extraction and optimization of algorithm for improving gait recognition rate.Although these methods achieve ideal gait recognition results under laboratory conditions,in real-time environment,the results are not so encouraging.In fact,there are many variables in the clinical environment,such as individual differences,human-computer differences and so on.Specifically,these variables include individual differences in different human physiological mechanisms.Even for the same person,under different load,different load style different speed,different road condition and fatigue,there are differences in physiological state,movement mode and mechanical characteristics.The main objective of this paper is to study whether these human-machine differences will have a significant impact on gait recognition,explore the ways to improve gait recognition rate at these levels of human-computer differences,and based on this research to develop real-time gait recognition system and a prototype of lower limb exoskeleton for pedestrian assistance.The main research work includes the following aspects:(1)Through literature research and related theoretical study,comb and grasp the frontier theory of gait recognition,and focusing on gait phase recognition,we study lower limb joints and degrees of freedom.According to the role of lower limb muscles in gait cycle,we select the target muscles to acquire electromyography signal for gait recognition.Combining with the mechanism of EMG signal generation,it lays a foundation for the research of gait recognition in multi-human environment and the development of real-time gait recognition system.(2)Through specific experiments,this paper explored in detail the effects of the variation of load and load style on gait recognition.In clinical settings,lower limb exoskeletons face many environmental variables,where we choose three common variables(load variation,load style variation,speed variation)to experiment.Overlapping analysis window and RMS,iEMG are employed to extract feature,and the traditional classification methods(BPNN,SVM,KNN)are chose to class.The Univariate multivariate analysis of variance was used to deal with the class result,and discuss the effect of the different load,load style,speed on gait recognition.(3)Through specific experiments,study how to use the auto feature extraction of CNN to Replace manual feature extraction such as RMS,iEMG,which usually lose feature.In this paper,we convert EMG signals into EMG images,with the help of the advantages of CNN in image processing and automatic feature extraction without losing feature to investigate the possibility of processing sparse EMG signals with CNN.The result showed that the ideal classification results can also be obtained by using CNN,through conversion of sparse EMG signal into two-dimensional gray image.(4)Development a real-time gait recognition system with EMG signal as input to realize real-time gait recognition from a practical point of view.In the system,we summarize the human-machine differences in real-time environment into four aspects,that are speed(3,5,7Km/h),load(10,20,30% weight of subject),load style(backpack,cross-shoulder,straight-shoulder),Slope of road surface(-15,0,15 degrees).These variables in system can be freely combined into human-machine variables in real-time environment,and with which to recognize gait phase.(5)The real-time gait recognition system is extended to a prototype of EMG lower extremity exoskeleton to perform joint angle prediction and control experiments.employ the pressure signal and joint angle signal as reference to label the EMG signa for gait recognition and prediction of joint angle.Then we predict the joint angle of knee and hip joints in two EMG signal source from different limbs and same limbs,Relative to the legs which using the lower limb exoskeleton prototype,to prove the validity of the real-time gait recognition system in Lower limb exoskeleton.