Research On Gesture Recognition Based On Surface Electromyography And Acceleration Sensor

The target object can be controlled by gesture recognition.The application of human-computer interaction based on gesture recognition in the fields of game entertainment and smart home has gradually integrated into our lives.On the other hand,as China is a large country with a disabled population,gesture recognition is of great significance to the harmonious integration of language dysfunction patients into the society.The convenience and accuracy of gesture recognition is the goal of researchers.Based on the existing research results in gesture recognition field,this paper takes the dynamic process of multiple gestures in Chinese sign language as the research object,and conducts an in-depth study on the classification of gestures by four channels of surface electromyography(s EMG)signals and one channel of acceleration(ACC)signal of the forearm.The main work and innovation of the paper are as follows:(1)Signal acquisition: Performed dynamic process analysis based on the form of gesture standard action,used a three-axis acceleration sensor fixed on the back of the wrist to collect acceleration signals so that could reflect the waving information of arm.From a physiological point of view,s EMG signals were collected from the forearm muscles of thumb extensors,finger extensors,radial wrist extensors,and superficial flexor flexors,which are closely related to finger or wrist movements,thus total of 5 channels of signals were collected for gesture recognition analysis.An experimental scheme was designed for the specific acquisition process of ACC and s EMG signals.(2)Signal preprocessing: An improved wavelet threshold denoising method was proposed to denoising the s EMG signals,and ACC signals were smoothed by the median filter method.After normalization of each channel signal,the threshold methods were used to extract the signal activity segment.(3)Segmentation method: Based on the analysis of the complex execution process of dynamic gestures,a segmentation method of complex dynamic gestures was proposed.The starting and end point of the signal activity segment of s EMG and ACC were used to decompose the complex dynamic gestures into three segments: starting segment,the main feature segment and the finishing segment,which are three relatively independent segments.The starting segment only contains ACC as the active state,and s EMG is the resting state,which reflects the macro information of ACC during the starting process of the gesture.The main feature segment includes information that both s EMG and ACC are active.This segment reflects the detailed information during the execution of the gesture.The relative change of ACC is accompanied by the rich information of s EMG,playing a decisive role in the recognition of gestures.In the finishing segment,s EMG shifts to the resting state,which reflects the detailed information of s EMG during the gesture termination process.(4)Feature extraction: According to the characteristics of ACC and s EMG signals of each segment,the macro and detail features of the three segments were extracted,and the macro information and micro detail features of dynamic gestures were organically combined to extract time domain feature combination,frequency feature combination,time-frequency feature combination,entropy feature combination and fusion domain feature combination.Constructed FACC,Fs EMG and FA+S feature combinations,PCA was used to reduce the dimension of high-dimensional feature vectors.(5)Gesture classification: Used a variety of classification methods to carry out a comparative analysis of gesture classification results from the segmentation idea and corresponding feature extraction method.The recognition results of KNN,ANN and SVM classifiers were compared.The feature vectors of s EMG + ACC mentioned in this paper achieved recognition accuracy above 90%,which verifies the effectiveness of this research method for dynamic gesture recognition.(6)Research and practice: developed an online sign language translation APP,effectively recognized 7 kinds of daily sign language,displaying text and broadcasting voice on the mobile phone by the time performing gestures,making the communication between deaf and dumb people and normal people more natural and fluent.