Research On Upper Limb Motion Sensing Technology

At present mainly in the field of upper limb power suits are rigid power suit and flexible dynamical suit,no matter at home and abroad,more than previous research focus in the field of the rigid power suit,and the rigid power suits usually have high quality weight,power,inflexible faults,flexible power suit mainly consists of flexible materials such as cloth,dupont line,with features of light weight,low consumption and good man-machine compatibility.At present,in the field of flexible power assisted clothing,the research is mainly on the lower limb flexible power assisted clothing,while the research on the upper limb flexible power assisted clothing is relatively few,and most of them are used for rehabilitation training,and in terms of power,the wearer's movement is relatively fast,so the requirements for real-time are higher.The research focus of this paper is to improve the real-time performance of the system while improving the accuracy of action classification.In this paper,the upper limb motion perception technology is mainly studied.Because surface EMG signal has the characteristics of easy acquisition,no damage to human body when collecting signals and rich information of human motion intention,it is used for motion intention identification.This paper mainly includes the following contents:This paper expounds the control scheme of the upper limb flexible power assist suit,and obtains three perceptual schemes of motion classification,stage classification of motion and Angle regression of upper limb.The anatomical structure of human upper limb bones and muscles was analyzed,and surface EMG signals of brachioradialis,biceps,triceps,deltoid and infraspinatus were collected by surface EMG sensors.Surface EMG signal was preprocessed.Since surface EMG signals belong to microsignal S and are prone to noise interference,Butterworth and Chebyshev filters are used for filtering,and the characteristic values of filtered signals are extracted.The motion classifier was designed by decision tree algorithm,quadric algorithm and support vector machine,and the upper limb motion was classified according to the characteristic value of the input surface EMG signal.The accuracy of the three kinds of classifier was up to 100%,and the three kinds of classification algorithms were compared from the accuracy and training speed.In the design of stage classifier of action,decision tree algorithm,quadric algorithm and support vector machine are used first,and then principal component analysis is used to process the feature.After finding that the classification effect is not ideal,neural network is used,and the classification accuracy reaches 95.2%.Finally,the accuracy of the upper limb Angle regression device is 99.3% by using the short and long time memory network.Finally,the real time of perception is analyzed.