Gait Pattern Segmentation And Gait Recognition Based On SEMG Signals

The lower limb gait is the combined action of the central nervous system,joints and muscles.Due to the periodic and diversity of human gaits,a complete gait cycle can be subdivided into multiple modes.Segmentation and identification of gait patterns have important applications in some research areas,such as intelligent prosthetic control,assisted rehabilitation,etc.The gait recognition is mainly focused on extracting multiple features and seeking suitable classification methods to achieve high classification accuracy.Based on the multi-source gait signals,the researches on gait segmentation and gait recognition were developed from the pre-processing signal to multiple feature extraction.The main contents are as follows:With the analysis of lower limb locomotion,the multi-source signals were acquired synchronously by the gait experiment system,which are the acceleration signals,surface electromyography(s EMG)signals and 3D motion capture signals.For the gait pattern segmentation,a simple algorithm was proposed to determine gait events in the dissertation.A complete gait cycle was preliminarily segmented using the comprehensive change rate of the acceleration signals.The heel and toe displacements were calculated by integrating twice in the frequency domain.According to the characters of foot displacements,four gait events were determined based on preliminary segmentation results,and validated by the 3D motion capture system.The results indicated that the algorithm has improved the accuracy of gait events and changed the layout of sensors compared with other methods.In order to reduce the influence of noise on gait recognition,the wavelet threshold method,the wavelet packet threshold method and the wavelet modulus maxima method were selected to delete noise.The signal to noise ratios and root mean square errors were calculated to evaluate noise elimination performance.After comparative analysis,the wavelet modulus maxima method behaved best for the de-noise processing.Based on the periodic,non-stationary and chaotic characteristics of the s EMG signal,the overlapped window method was proposed to calculate the time-domain and frequency-domain features.The results showed that the time-domain and frequency-domain features had large difference among the different gait phases.The approximate entropy,sample entropy and fuzzy entropy were extracted respectively.The fuzzy entropy behaved high discrimination through the comparative analysis.For acquiring high recognition accuracy and efficiency,the support vector machine(SVM),the extreme learning machine(ELM),and linear discrimination analysis(LDA)were applied to classify different gait patterns.Then the influencing factors of gait recognition were analyzed from the classification accuracy,which included the feature window length,the window increment,the feature dimension,and individual differences and sample sizes.These factors have a certain impact on the stability and efficiency of the gait identification.The gait pattern segmentation and recognition were studied systematically in the dissertation,which included gait event determination,s EMG signal denoising,multi-feature extraction and gait classification.The recognition accuracy and stability were improved by the analysis of some influencing factors.