Research On Complex Gait Recognition And Stride Length Estimation Based On SEMG Signals Of Human Lower Limbs

The surface Electromyography(sEMG)signal is a weak bioelectric signal detected on human skin surface.It has the ability to represent human motion information.The sEMG signals and human motion information can be modeled by pattern recognition algorithm,so that the motion patterns and stride length information of human motion can be obtained.These human motion information play an important role in assisting pedestrian navigation and positioning(PNP).PNP technology is an increasingly important location information sensing technology in human life.It is widely used in Navigation technique of Individual soldier,Unknown area exploration,Indoor fire protection,Space exploration and so on.Currently,the application of sEMG signals in PNP is still in the exploratory stage.The gait patterns studied are mainly focused on straight walking,upstairs and downstairs.The recognition of gait motions are mainly the study of separate gait pattern and gait subdivision mode.There are few studies on mixed gait,complex gait and turning gait.Therefore,this paper carries out the research of using sEMG signals to identify human lower limb gait motions,expands more patterns of human gait,and studies the multi-gait recognition of mixed gait mode.In addition,the human turning gait is also explored.Finally,the stride length training is carried out for more gait patterns.The main research works include the following aspects:(1)A four-channel sEMG signal acquisition system was built.The sEMG signals of Gastrocnemius medials,Vastus rectus,Vastus lateralis,and Gluteus medius of lower limb were acquired.Separate gait pattern and mixed gait patterns experiments were conducted to acquire the sEMG signals of six gait patterns,such as walking straight,going upstairs,going downstairs,running,jumping,and turning.(2)A new sEMG signal filtering algorithm,BFAWT,is proposed,which combines band-pass filtering with adaptive wavelet threshold.This method has good adaptability to the sEMG signals of various muscles and gait patterns,and can effectively decrease baseline wandering,artifact noise and Gaussian noise of raw sEMG signals.An active segment detection method based on the peak energy of sEMG signals is designed to improve the poor adaptability of the classical double threshold method(DT method)to different sEMG signals.(3)Two recognition algorithm models,support vector machine and BP neural network,are constructed.Based on the single gait pattern experiment,the sEMG signal performance of different muscles and the combination of feature vectors are analyzed.The recognition rate of multi-gait patterns under complex gait conditions is studied by hybrid gait experiment.Compared with previous studies,more gaits are studied.From the results,under the complex gait conditions,the average recognition rate of the six gaits is 88.36%.In addition,this paper explores the human turning gait.By designing the turning gait experiment,we found the difference in the time sequence between the walking and the turning gait of the sEMG signals.The difference is introduced as a feature to identify the turning gait,and the recognition rate reaches 99.53%.Finally,the prediction of stride length is studied,On the basis of straight walking stride length estimation,turning stride length estimation is studied.In the test of 16 sets of walking gait experimental data,the estimation error of stride length was less than 0.12 m,and the average estimation error of accumulated 10 strides was less than 0.06 m.In 15 sets of turning gait stride length experiments,the maximum error of stride length estimation is less than 1.7%,and the error is less than 0.02 m.The research in this paper verifies that the complex gait pattern recognition,and stride length estimation based on sEMG signals have certain accuracy,and provide reference for the study of pedestrian autonomous positioning technology.