Master thesis presentation by Hans Emil Atlason at MPBME: Inertial sensors for improving electromyographic classification of hand-gestures

On Wednesday, 1 June 1, Hans Emil Atlason, MPBME, will present his master thesis with the title “Inertial sensors for improving electromyographic classification of hand-gestures”.

Time: 14.00 pm. 1 June, 2016

Place: Landahlsrummet (room 7430), Hörsalsvägen 11, 7th floor

 

Examiner: Bo Håkansson

Supervisor: Max Ortiz Catalan

Welcome!

Abstract
Introduction: Classification of hand-gestures from electromyography (EMG) could provide robotic prosthetic hand control to transradial amputees. However, the usability of such a system is influenced by many factors, such as arm position and limb movements. Inertial sensors can be used to provide information about arm position and movement, and therefore potentially improve prosthetic hand control. The aims of this thesis were to implement the technology to simultaneously record forearm EMG and inertial signals for hand-gesture classification, evaluate recorded EMG and inertial signals during various movements, and conduct an experiment to compare offline and real-time classification of hand-gestures using different feature-sets of EMG and inertial signals.

Methods: A Motion Processing Unit (MPU) was used to provide accelerometer, gyroscope and magnetometer data. Software for inertial signal recordings was developed in Matlab and added to the myoelectric pattern recognition software BioPatRec for movement classification. Prototype EMG and inertial signals were recorded while a subject performed sharp up-and-down arm movements, and contraction and relaxation of a hand gesture. In the comparison experiment, subjects performed hand gestures while forearm EMG and inertial data was recorded. Signal features were then extracted from EMG, accelerometer, gyroscope, and orientation in Euler angles. Classification performance was compared offline and in real-time (Motion Test).

Results: A Graphical User Interface (GUI) was developed for recording of EMG and inertial signals, and modules for inertial signal treatment were added to BioPatRec. The prototype recordings of simultaneous EMG and inertial signals showed that changes in inertial signals due to arm movements can correlate to movement artifacts in EMG, and that the contraction and relaxation of a hand gesture can be traced using inertial signals. Offline classification showed no statistical differences between the EMG alone and EMG with either accelerometer, gyroscope or Euler angle features. The accelerometer and Euler angle features resulted in higher accuracy than gyroscope features, despite that inertial signals alone are not intended to distinguishing between hand-gestures. No statistically significant difference was found between EMG alone and EMG with gyroscope in the real-time classification of hand gestures.

Conclusions: Inertial sensors showed to provide complementary information to EMG during the execution of hand gesture. This work provides a platform to investigate the potential benefit of inertial sensors for prosthetic control.