Human-Computer Interaction Method For Exoskeleton Robot

Nowadays,with the increased incidence of cardiovascular and cere-brovascular diseases and the problem of aging population,they create more and more serious social problems on one hand,and make a kind of new technology gradually become a focus of emerging research on the other hand.Exoskeleton robot is a kind of intelligent robots,which can be worn by a person to assist and protect the wearer.It is developed based on the man—machine system,which can not only enhance human movement capabilities,but also obtain the limb's information effectively and precisely in order to control the robot.Perceiving behaviors with sensors is an important part of exoskeleton robot research.However,the researches about human movement intention recognition at home and abroad still mainly focus on hand and upper limb up to now,lacking the studies aimed at lower limb.In this thesis,the sEMG and acceleration signals of lower limbs from normal people and the disabled are collected by using a wireless Delsys collection system.Lower limb movement pattern classification is realized by processing the signals and the results can further be used for the research of exoskeleton robot.The main work of this thesis is as follows:Firstly,this thesis contrasts the research status at home and abroad of exoskeleton robot and analyzes the human lower limb motion mechanism.According to the requirements of the lower limb exoskeleton system,it designs suitable reference for the lower limb exoskeleton robot and introduces system components of lower limb exoskeleton robot which is designed by the author.Secondly,this thesis determines the experiment scheme and collects lower limb movement data.According to the comparison between sEMG and acceleration signals and different lower-limb motion modes,the author draws the conclusion that different lower-limb motions have different signal excitation time and excitation grade.Then,it preprocesses the original signal data and extracts the different signal features.Later,it selects the steady feature and makes the PCA dimension reduction of the extracted signals,and provides the necessary conditions for the back of the classification and recognition.After that,it identifies the classification of lower limb movements by using linear classifier,artificial neural network and support vector machine respectively.Lastly,it compare the differences of the classification results of different identification method,and gets the most suitable identification method for sEMG and acceleration signal of lower limb motion.Finally,it compares the normal lower limb movement signals and disables lower limb movement signals.According to the disabled signal characteristics,the author chooses the signals which have better quality by using the method improved in this thesis.Then,it completes the wavelet noise of selected data with the improved threshold de-noising method in this thesis.And then identifies the classification with the de-noising data and gets a better classification results.Lastly,it compares the results to demonstrate the validity of the data selection and improved threshold method.This thesis achieved the characteristics fusion of sEMG and acceleration signals.The correct recognition rate of normal lower limb movements can reach 95%,and the correct recognition rate of disabled lower limb movements can reach 80%,which can lay a certain foundation for follow-up study.