A human-like dexterous EMG prosthetic hand control method based on gesture recognition

Abstract

The invention discloses a human-like dexterous EMG artificial hand control method based on gesture recognition, which recognizes eight kinds of gestures of a user in real time and operates the dexterous artificial hand to carry out synchronous action. The gesture recognition strategy of this control method is based on neural network algorithm, participants first completed eight preset gestures (relaxation, wrist valgus, wrist varus, fist-clenching, palm-extension, gesture 2, gesture 3 and thumb-upright) during the training phase, and then the system is able to recognize any one of the eight gestures randomly completed by the user in real time. The method uses a Tensorflow machine learning framework to learn weights and perform visual analysis. The method collects, trains and predicts the surface electromyography signal of a user, the comprehensive prediction accuracy of the 8 gestures reaches 97%, and the training is not needed when the gesture is worn again. When the subjects actuallycontrolled the artificial hand, the voting algorithm was used to optimize the real-time gesture prediction, and the synchronization rate of the artificial hand reached 99%.

Description

technical field [0001] The invention relates to the technical field of gesture recognition, in particular to a control method for a humanoid dexterous myoelectric prosthetic hand based on gesture recognition. Background technique [0002] The recognition method of action intention by using bioelectrical signal mainly includes the recognition based on EEG technology and the recognition based on EMG technology. Although EEG technology can directly obtain more action intention information from the brain, EMG technology currently has a higher application prospect due to its natural and comfortable interaction method and more stable data acquisition and processing method than EEG. [0003] Early domestic research on electromyographic signals mainly used multiple electrodes to detect specific muscle tissues. The detection objects mainly included flexor carpi ulnaris, extensor carpi ulnaris, and finger extensors. The early methods of gesture recognition were time-domain analysis, ...

AI Technical Summary

Problems solved by technology

[0004] Based on domestic and foreign research results, it is found that compared with monitoring the EMG signals of specific muscle tissues, the array EMG monitoring method for the entire forearm muscle group can reproduce more abundant hand movement intentions, but the channel of the array EMG monitoring Excessive numbers will not only lead to excessive redundancy of signals and increase the difficulty of processing, but also the device responsible for EMG collection will be larger and more complex, reducing the comfort of using the gesture recognition system

Images