Muti-posture gesture recognition based on a two-hand signal parallel network

• 1、School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing 100876

Abstract：Hand gesturerecognitionbasedonsurfaceelectromyography(sEMG) signalhasshownbroadapplicationprospectsinfieldssuchashuman-computerinteractionandrehabilitationmedicine due to its safetyandconvenience recently.A large number of studies have been conducted on hand gesture recognition using sEMG. However, most existing studieswere carried out under ideal laboratory conditions, neglecting the impact of body posture on recognition models for hand gesture recognition. This work carrys out multi-gesture recognition under various postures based on surface electromyography signal, designing a two-hand signal parallel network for hand gesture recognition under multi-postures. Results indicate that this method achieves goodhand gesture classification performance on different postures and when all postures are confused, with an average accuracy of about 89.7%, which is about 7.3% higher than the singel hand network. Besides, it performs better in scenarios with all postures confused and limited training data, with an improvement up to 9.7%. It demonstrate that the two-hand signal parallel network has stronger robustness when doing hand gesture recognition under various postures.

Keywords： Artificial intelligence Surface electromyography signal Body posture Hand gesture recognition