Ankle Joint Action Recognition Of Multi-feature Fusion And Virtual Simulation

Human action recognition has been widely concerned by many researchers as the key technology of human-machine interaction,mobile robot and rehabilitation engineering.At present,the signals of action recognition mainly include physiological signals,biomechanical signals and video image signals.EMG,as an important physiological signal of human body,reflects the state of neuromuscular activity.As an important biological force signal,joint acceleration signals contain abundant information on human joint activity.Therefore,this dissertation focuses on the research of synchronous acquisition,preprocessing,feature extraction,action recognition and virtual simulation of surface EMG and ankle acceleration signal.The specific work is as follows:(1)Surface EMG signal and joint ACC signal are collected in real time.Design experimental paradigm,synchronous acquisition the surface EMG signals of four muscles including tibialis anterior muscle,long peroneal muscle,gastrocnemius muscle and tibialis posterior muscle and the three axis ACC signals of ankle joint through the DELSYS wireless electrode surface EMG collection system when ankle joints perform four actions containing toe flexion,dorsiflexion,foot varus and foot eversion.(2)Preprocessing of the original EMG signal and ACC signal.Design a method combining adaptive filtering with wavelet denoising,and effectively remove high-frequency noise,motion artifacts,and 50 Hz power frequency interference in EMG signal.Filter out noise and interference in ACC Signal using wavelet threshold denoising.Finally get pure EMG signals and joint ACC signals.(3)Feature extraction of surface EMG signal and ACC signal.Extract integral EMG,RMS value,average power frequency and median frequency of the EMG signals in the time domain and frequency domain respectively.Apply a nonlinear analysis method to extract the nonlinear features of the EMG signal such as fuzzy entropy,approximate entropy and sample entropy.At the same,calculate time domain features of the joint ACC signal including absolute integral mean value,variance,correlation coefficient,amplitude peak and amplitude mean.At last,analyze the clustering characteristics of various features about four action modes by statistical scatterplot.(4)Feature fusion and classification recognition.First,we can input the time domain features,frequency domain features and nonlinear features into the radial basis function support vector machine separately to achieve action classification on a single feature.Then,according to the decision results of classifier to weight features,we can obtain the fused feature vector which is put into support vector machine for final classification recognition.The average accuracy rate of ankle motion recognition is 95.83%,which is higher than the recognition accuracy of single feature vector on EMG signal or acceleration signal.The result of simulation experiment shows that the method to multi-feature fusion can significantly increase the classification accuracy of the classifier and improve the precision of ankle joint motion recognition.(5)Virtual model design and simulation realization.First of all,we can import the action recognition results of classifiers into LabVIEW software through data communication interface between MATLAB and LabVIEW.In the next place,we designed a virtual lower limb three-dimensional model by SolidWorks software to achieve four kinds of action simulation of lower limb ankle joint which is toe flexion,dorsiflexion,foot varus and foot eversion and imported the action simulation into LabVIEW software through a fixed path.In the end,we designed the Greedy snake move game through the LabVIEW software and controlled the moving direction of the snake by four movements of the ankle.It can provide rehabilitation training with interesting and experience feeling through virtual games.The research results of this dissertation can be applied to the field of human-computer interaction rehabilitative training,and have dual values of science and application.