Gait Recognition Method Based On Surface Electromyography And Acceleration Fusion Of Rectus Femoris

Only correctly identify the movement pattern of lower limb amputees,in order to take effective control strategy,therefore,lower limb gait pattern recognition as one of the key technologies of intelligent prostheses.Aiming at the complex multi-sensor system in gait recognition,high cost,redundant information,poor user wear experience,high correlation between information,difficulty in feature engineering,and difficulty in establishing classification model,a simple and low cost method is proposed.A gait recognition method based on the myoelectric-acceleration fusion of the rectus femoris surface.Based on signal analysis and pattern recognition technology,the five common gait of lower limbs walking,up stairs,down stairs,uphill and downhill are accurately identified.The main work contents are as follows:Firstly,according to the advantages and disadvantages and internal relations of commonly used information sources in gait recognition,surface electromyography and acceleration signals are selected to identify the lower limb asynchronous mode;in order to collect signals,a wireless signal acquisition system is built,and the myoelectric acquisition circuit is completed.Design,acceleration sensor,Bluetooth module and microcontroller selection,developed a wireless communication protocol for data transmission and reception,and programmed a single-chip C language program and a LabView-based PC program.Secondly,it explains the main muscles of the lower limbs,as well as the position and function of these muscles.According to the position,area,length,intensity of contraction,cross-talk,and interference of signal acquisition,the rectus muscles are selected to collect surface myoelectricity.And acceleration signals;subsequently developed the experimental flow and specifications of signal acquisition,collecting surface EMG-acceleration signals in five gait modes: flat walking,up stairs,down stairs,uphill and downhill;Noise,improve signal-to-noise ratio,surface EMG signal denoising uses second-order Butterworth filter,acceleration signal uses wavelet threshold to denoise,and determines the starting moment of motion based on sample entropy.Then,extract the data of 200 ms after the starting point of the action,extract the mean value of the time domain eigenvalue,the root mean square,the standard deviation,the mean absolute deviation,the interquartile range,the correlation coefficient,the average power frequency of the frequency domain eigenvalue,and use the random forest.The importance scores of different eigenvalues are calculated,and the correlation coefficients between the features are calculated to complete the preliminary screening of the eigenvalues.Finally,according to the eigenvalue importance score from high to low,the feature input is gradually added to the training classification model in the XGBoost algorithm,and the eigenvalues with little influence on the recognition result are removed.After the KPCA transformation,the final classification result is obtained,and respectively,in a single signal.Compared with the recognition results of source and surface EMG-acceleration fusion and other classification algorithms,the recognition rate of this method is the highest,which indicates the rationality of information source selection and the effectiveness of classification algorithm selection.