Research On Multi-sensor Human Motion Recognition For Upper Limb Exoskeleto

In recent years,with the aggravation of the population aging,the number of hemiplegic patients caused by stroke is increasing.In order to make more hemiplegic patients get scientific and reasonable rehabilitation training,many scholars at home and abroad have devoted themselves to the research of rehabilitation robots.However,most of the existing rehabilitation robot systems drive patients to carry out repetitive and mechanical passive training,which unable to identify the patients’ exercise intentions,the patients’ participation in it is low,so it is difficult to mobilize the enthusiasm of rehabilitation training.In order to solve this problem,aiming at the upper limb motor function rehabilitation of hemiplegic patients,this paper designed and built a multi-sensor motion recognition system based on sEMG signals and limb posture and motion information,and used the designed upper limb rehabilitation exoskeleton platform to complete the online application of the motion recognition system.The main work is as follows:Firstly,based on sEMG and IMU signals,the human upper limb motion sensing system is designed and built.Combined with the upper limb rehabilitation training of hemiplegic patients,eight kinds of upper limb movement training actions and the arrangement positions of two kinds of sensors are determined,and the sensing collection experiment is designed to complete the sampling of human upper limb sensing data.Secondly,the sensing data are preprocessed and feature extracted.Butterworth filter and comb filter are used to filter and denoise two kinds of sensing signals.Based on short-term energy and short-term variance,the active segment detection of the signal is completed,and the threshold adaptive setting in the detection process is realized by using the maximum probability.Then,the representative feature values of the two kinds of sensing signals are extracted,and48-dimensional features of s MEG signals and 24-dimensional features of IMU signals are obtained respectively.Thirdly,based on the machine learning algorithm,the feature analysis of sEMG and IMU data of human upper limbs is carried out,and a feature processing method combining LDA and DCA is proposed,and the feature reduction and fusion analysis of the two types of sensor data are carried out in turn.SVM classification algorithm is used to identify and classify the upper limb movements of the fused signals,and the accuracy rate reaches 99.67%,which is significantly higher than that of sEMG or IMU signals alone.Finally,an upper limb rehabilitation exoskeleton test platform is built to verify the multisensor motion recognition model constructed in this paper online,and the average correct recognition rate of eight kinds of upper limb motions is 95.17%,which verifies the effectiveness and feasibility of the multi-sensor motion recognition system based on sEMG and IMU signals in this paper,and can effectively improve patients’ active participation in rehabilitation training.