Research And Implementation On Action Recognition Of Forearm Prosthesis Based On EMG

Forearm disability has a negative effect on the physical functions and labor ability for the disabled persons who lose forearms or hands.With the help of forearm prosthetics,partial limb functions are recovered through engineering methods.In recent years,due to the development of science and technology,increasing number of high precision prostheses are available in the medical products market,including prosthetic products which can directly control a single finger.Collecting surface electromyography(sEMG)signal through the electrode interface and generating control instructions,these prostheses make it possible for the disabled to control the artificial limbs to complete some specific actions.However,high precision products are usually with high price,while the laboratory electromyography method,mainly tested on the computer simulation platform,cannot be applied to practical use temporarily.To solve the above problems,this paper attempts to design a set of EMG signal acquisition and motion recognition system for intelligent prostheses,based on an open source embedded platform.This system,aiming at high precision intelligent prostheses,is designed to achieve forearm surface EMG signal acquisition,action recognition and control signal generation.This design mainly includes the following aspects:1.The current schemes of EMG signal acquisition and preprocessing are studied,as the whole investigation process is based on muscle movement units distribution and EMG signal generation principle.During this period,surface EMG signal acquisition system,electrode position arrangement and digital signal preprocessing method are designed.2.Based on the pattern recognition technology,the related feature models and their methods of feature extraction in surface EMG signal processing are introduced.The method of signal feature extraction with integrated time domain and autoregressive feature is analyzed and compared with another method of domain and power spectrum description.The advantages and disadvantages of each method are analyzed through experiments.3.On the basis of existing methods,a new method of EMG signal processing and pattern recognition suitable for embedded system is proposed by combining finite state machine and pattern recognition methods: FSM-TSD.This method split a large number of classification problems according to different states,which reduces the classification difficulty and improves the classification accuracy.4.A set of EMG signal acquisition and motion recognition system scheme is proposed to achieve a balance between acquisition performance and market parameters,according to the requirements of commercialization of EMG prostheses control system.Furthermore,implementation methods of various algorithms on resource-limited embedded micro-controller platform are reviewed in this paper.The experiments of EMG signal acquisition and motion recognition,using the embedded platform,are carried out as well.