Myoelectric Signal Processing for Prosthesis Control

von David Hofmann
Dissertation
Datum der mündl. Prüfung:2014-02-05
Erschienen:2015-02-03
Betreuer:Prof. Dr. Florentin Wörgötter
Gutachter:Prof. Dr. Florentin Wörgötter
Gutachter:Prof. Dr. Dr. Dario Farina
Zum Verlinken/Zitieren: http://dx.doi.org/10.53846/goediss-4905

 

 

Dateien

Name:Thesis_DH.pdf
Size:33.3Mb
Format:PDF

Lizenzbestimmungen:

Zusammenfassung

Englisch

The myoelectric signal (MES) is the electrical manifestation of contraction. The MES, recorded at the surface of the skin, has been exploited as a control source for powered upper limb prosthetics. The primary objective of this work is to improve control strategies based upon pattern recognition algorithms. We first deal with the MES amplitude estimation problem and show that a Bayesian filter, suited for non-stationary signals, leads to subtantial improvement over common amplitude estimators. Based on high density electrode array recordings we then compare data dependent to data independent spatial filters and provide evidence that second order blind source separation is practical when transient signal regimes are included. Finally we address the problem of electrode selection. Here we compare different selection algorithms and provide means to find good electrode placement locations with respect to classification accuracy.
Keywords: machine learning, myoelectric signals, pattern recognition, amplitude estimation, Bayes filter
 
Statistik