Wireless sEMG-based identification in a virtual reality environment

Abstract

A VR real-time control system based on wireless sEMG was designed. The scene in VR was the kitchen at home. The system included sEMG acquisition module, software control module and VR environment module. The sEMG signals were collected from the subjects by a portable wireless acquisition module. The software control module consisted of four parts: real-time myoelectricity signal detection and segmentation, feature extraction, classification identification and control instruction. The mean square and moving average window method were used to segment sEMG signals. The mean absolute value and singular values of wavelet coefficients were selected as sEMG features. Support vector machine and probabilistic neural network were applied for model training and classification. Finally, the control of four motions in virtual kitchen was completed. The VR environment module was derived to perform different motions according to instructions provided by the classification in virtual kitchen.The experimental results showed that this system could perform the real-time control of virtual kitchen motion. For myoelectricity signals of standard motions, the average offline identification accuracy was 95%, and the average online identification accuracy was 90.31%. For the ones of real motions, the offline accuracy of 86.09% and the online accuracy of 83.33% were achieved respectively. This system can be used for muscle rehabilitation training, and provide immersive virtual kitchen scene, which is of positive significance to the rehabilitation of patients. It can also provide an assessment analysis of the muscle recovery of patients.