Research On Gesture Recognition Method Based On SEMG Signal And Acceleration Signal

Today,there are 650 million disabled people in the world.The number of disabled people in China has reached 80 million,of which the number of physically disabled people accounts for about42%.As two representative perceptible signals,surface electromyography(s EMG)signal and acceleration signal can be used for human motion recognition,so as to help the disabled control prosthetic devices.Therefore,this thesis mainly studies the method of action recognition based on s EMG and acceleration signal,and designs and implements a real-time action recognition systemFirst of all,this thesis proposes an action recognition algorithm based on multi feature fusion,which mainly includes time sequence signal filtering and segmentation,multi feature extraction and model recognition.It extracts a variety of time-domain,frequency-domain and other types of features of s EMG and acceleration signal,and fuses them through tree model to achieve the purpose of identifying the action represented by the signal.After the experiment,we get the best recognition accuracy of 79.79% in the case of only surface EMG signal input,and 93.38% in the case of additional acceleration signal input.The model can effectively solve the problem of low recognition accuracy of traditional recognition methods.Then,an end-to-end action recognition method is proposed.In the input phase,the method does not need to mine the artificial features,but only needs to segment and normalize the signal to input the network.Moreover,our model has a certain degree of scalability.We can dynamically adjust the number of feature extraction modules and channels as well as the types of input signals according to the application scenarios of the model,so as to achieve the effect of balancing accuracy and reasoning speed.The experimental results show that the accuracy of the model can reach 81.76% when only s EMG data is input,and 94.65% when mixed s EMG and acceleration data is input.Moreover,the reasoning speed of the model on CPU is only 38.6ms.It effectively solves the problems of tedious manual feature mining and difficult model deployment at the edge.Finally,this thesis designs and implements a real-time action recognition system,which mainly includes two parts: acquisition device and host computer system.In the part of acquisition equipment,the design and implementation of hardware and software are introduced.In the PC part,the design and implementation of data acquisition PC and action reasoning PC are introduced.The neural network recognition model is deployed in the upper computer of action reasoning,which can complete the real-time detection of action and display the angle information of human arm in real time.The delay can be controlled within 200 ms,and the accuracy is more than 90%,which basically meets the requirements of real-time recognition.