sEMG-based continuous hand gesture recognition using GMM-HMM and threshold model

To detect the person's intention for the control of upper-limb exoskeleton robot, we propose a recognition frame of continuous hand gestures. This frame is mainly concentrated on dynamic segmentation and real time gesture recognition based on sEMG. The hand gesture was modeled and decomposed by the use of Gaussian Mixture Model-Hidden Markov Models (GMM-HMM). GMMs are employed as a sub-states of HMMs to decode sEMG feature of gesture. The log-likelihood threshold and KL-divergence threshold are adopted to select target gesture model. In myoelectric control schemes, the sEMG data are collected by Myo armband 8-channels sEMG sensors. The proposed framework has ideal classification accuracy and its simpler acquisition armband make it attractive to a real-time myoelectric control system.