Study On Motion Intention Recognition Of Human Knee Based On Mechanomyography And CNN-SVM Model

With the aggravation of population aging process,the number of the elderly is growing rapidly,and solving the problems of inconvenience in daily life caused by the weakening of limbs movement ability has become an urgent matter for the elderly.On the other hand,the number of amputees is increasing year by year due to vascular diseases and accidental injuries,and the targeted rehabilitation care and training are needed for the amputees.But due to their children being busy with work,insufficient nurses,rising labor costs,etc.,caring for the elderly and the physically disabled has become a thorny problem for every family.The wearable power-assisted robots or powered prostheses can provide movement assistance and rehabilitation training for the elderly with weakened limbs movement ability or the physically disabled,and improve their self-care ability and life quality,thereby reducing the burden of manual care and even no need of manual care.With the development of human-machine interaction technology,the wearable power-assisted robots will gradually turn from accepting instructions passively to recognizing and understanding the human motion intention actively.The robots can follow and assist the human motion in real time and appropriately according to the human motion intention,thus realizing the human-machine coordinated motion.Therefore,the human motion intention recognition is crucially important for the human-machine coordinated motion control.However,there are still some problems in the study of how to obtain the human motion information stably and conveniently and how to recognize human motion intention in real time and continuously.As a result,it is difficult for the robot to follow the human motion in real time and provide assistance timely,which affects the naturality and flexibility of human-machine coordinated motion.This dissertation takes the wearable power-assisted robots which are applied to the related fields of helping the elderly and the disabled and rehabilitation training as the research background,takes solving the key problems of human motion information obtaining and human motion intention recognition in the human-machine coordinated motion control technology as the research target,and takes the knee motion as the research object to explore a new method of human motion intention recognition.The main work and innovative achievements are summarized as follows:(1)A method of obtaining human motion information based on the mechanomyography(MMG)signal detected without direct skin contact was developed,which can detect motion information more minimally intervening in human body than using surface electromyography(sEMG)signal.This dissertation selected four muscles related to the knee motion according to the kinesiology and anatomy.In the experiment,MMG sensors were placed on the clothes against the selected muscles to detect multi-channel MMG signals,and the human motion information was extracted from the multi-channel MMG signals detected on clothes for the pattern recognition and angle estimation of knee motion.The experimental results verified the feasibility of obtaining human motion information based on MMG signals.This method has the advantages over the method based on sEMG signal to obtain human motion information.It will promote the research and development of wearable motion information detection devices,which are more stable,more convenient and more minimally intervening in human body,thereby greatly improving the flexibility,comfortableness and wearable ability of wearable power-assisted devices.(2)A method of human motion pattern recognition based on the MMG signal and the combined model of convolutional neural network(CNN)and support vector machine(SVM)was proposed,which introduced the CNN mainly applied to image processing into the direct processing of time series signal.This method overcame the drawbacks of hand-crafted features used by traditional classifier algorithms and improved the classification performance.In this dissertation,the four-channel MMG signals detected on clothes in knee motion experiment were used as the experimental data,and they were input into the CNN-SVM model in the form of time series signals to recognize the six knee motions.The model was evaluated by the evaluation methods of classifier performance,such as cross validation,confusion matrix,and receiver operating characteristic curve.The experimental results show that the CNN-SVM motion recognition model can not only automatically extract the effective features by means of the CNN's convolution and pooling operations,but also further improve the generalization ability and recognition rate of the model by means of the SVM classifier to process and classify the automatically extracted features,and has better effect compared with the traditional classification method using hand-crafted features.(3)A method of human motion estimation based on the MMG signal and the CNN-SVM model was proposed,which introduced the CNN applied to pattern recognition into the regression estimation of time series signal.The human motion angle estimation was achieved without the need of establishing complicated muscle model and handcrafting features.In this dissertation,the three-channel MMG signals detected on clothes in knee motion experiment were used as the experimental data,and they were input into the CNN-SVM model in the form of time series signals to estimate the knee motion angle.The model was evaluated by the evaluation indexes of regression model performance,such as root mean square error and correlation coefficient.The experimental results show that CNN-SVM angle estimation model can automatically extract effective features through the CNN,and further reduce the estimation error and improve the generalization ability of the model through the SVM regression.(4)A simulation environment for the application of human motion intention recognition was established,which verified and demonstrated the feasibility of proposed methods in the control of wearable power-assisted robot.This dissertation used the SolidWorks to create a three-dimensional virtual model of the human body and power-assisted robot leg.Under the combined environment of Lab VIEW and MATLAB,the method of knee motion intention recognition based on the motion recognition and motion angle estimation of knee joint was applied to the simulation control of the virtual power-assisted robot leg.The process of human-machine coordinated motion could be visually and intuitively observed,which facilitated the verification of the research methods.