A Comprehensive Evaluation of Hidden Markov Model for Hand Movement Recognition with Surface Electromyography

Surface Electromyography (sEMG) gesture recognition plays an important role in developing Muscle-Computer Interface (MCI) system and myoelectric controlled assistive devices. Its recognition accuracy is greatly affected by the selection of classifiers. This paper evaluates the performance of Hidden Markov Model (HMM)-based sEMG hand gesture recognition on the large scale Non-Invasive Adaptive Hand Prosthetic (NinaPro) Database. We conduct an HMM-based hand gesture recognition framework using sEMG signal and make comprehensive evaluations of three HMM classifiers (HMM with Gaussian emission (Gaussian-HMM), HMM with Gaussian Mixture Model (GMM-HMM) and Semi-Continuous-HMM (SCHMM)) using the Within-Subject cross-validation (WSCV). Our evaluation is based on three commonly used feature sets, and the experiments are conducted on seven benchmark databases of the NinaPro Database. The experimental results on the whole NinaPro Database show that SCHMM classifier consistently achieves the best performance among the evaluated HMM classifiers. This work presents comprehensive evaluation of three commonly used HMM classifiers on the benchmark database NinaPro and also proposes a new feature sets during the evaluation.