Research On Gait Recognition Method From SEMG Signals Of Lower Limb

Lower-limb wearable exoskeleton plays an important role in the fields of medical rehabilitation,national defense industry and so on.More and more researchers and engineers have begun to devote themselves to the development of intelligent assistive devices.Gait recognition is the key research content in this field.SEMG signals refer to the collective electric signal from muscles,which is produced during muscle contraction.Thus,the SEMG signals of lower limb can reflect the motion intention of body which contributes to the intelligent control of assistive device.The recognition system proposed by this thesis recognizes the gait stages according to the information reflected by SEMG signals.In view of the whole recognition process,this thesis carries out the following research:(1)The SEMG signals of lower limb is nonlinear,non-stationary and easy to be interfered.It is very important to choose an effective method to de-noise the signals in preprocessing.In the thesis,an improved method to select the threshold and threshold function of wavelet de-noising method is proposed.The experimental results show that the improved threshold and threshold function can significantly raise the signal-tonoise ratio and reduce the root mean square deviation of the processed signals.(2)In the feature extraction process,only when the starting point of the active segment is detected accurately,can the useful information be extracted at the appropriate position of signals.The thesis studies the traditional detection method based statistical characteristics or TEKO,then introduces the sample entropy which reflects the complexity of the signal sequence into the detection method,and finally achieves a more accurate detection of the starting point of the active segment.And then,the time-domain,frequency-domain and time-frequency-domain features are extracted from the active segments.These features are then evaluated and the features which are complex in calculation and useless in distinguishing gait stages are discarded.Then,the principal component analysis method is used to reduce the dimension of eigenvector.(3)How to recognize each gait phases also a multi classification problem.The thesis focuses on the support vector machine(SVM)and linear discriminant analysis(LDA).SVM is suitable for non-linear classification,but its training time is long,LDA is useful in linear classification,and the computational complexity is low.The thesis combines the two models to recognize the gaits through directed acyclic graph.The experiment results show that the combined model which is effective in multi classification includes the advantages of high classification accuracy in SVM and short training time in LDA.This thesis focuses on how to get key information from SEMG signals and how to use the information to recognize gaits accurately and quickly,and completes signal denoising,signal feature extraction and recognition model construction.By analyzing the influence of model parameters on the classification results,appropriate parameters are selected to improve the accuracy and speed of the model.