Advanced Processing of sEMG Signals for User Independent Gesture Recognition

Abstract

While the classification of gestures recorded with sEMG can reach very high recognition rates when the user has trained on the system, performance obtained on unknown users remains low. In this work we attempt to use advanced signal processing and pattern classification methods for improving classification performance of gestures on unknown users. Our approach is to take an existing feature set, add promising features, and use feature selection to prune poor features. For classification we use a support vector machine with a Pearson VII kernel, for which a particle swarm optimization was used to search through its parameter space. Results are presented on the NinaPro database, and show excellent results when the user is known to the system as well as a significant improvement on existing work when the user is unknown.