Real-time embedded EMG analysis for human-computer interaction

Abstract

Recent times have seen the increasing use of biophysical signals in numerous Human Computer Interaction (HCI) applications. Electromyographic (EMG) signals fall into this class of phenomena and are generally sensed at the skin’s surface. EMG signal analysis is popularly used for classifying a wide range of wrist-hand movements within prosthetic and gesture control equipment. To enable very accurate detection of such movements, high complexity EMG signal analysis methods are used which demand powerful computing resources. For portable systems, such as prosthetic limbs, real-time low-power operation on hardware platforms is critical. Several improvements have been carried out to accurately detect and classify a wide range of wrist-hand movements, but still have issues such as precise sensing, real-time hardware processing, and security. To address these challenges, this thesis proposes a novel approach for detecting wrist-hand movements with low-complexity EMG signal analysis system executes in real-time. A novel biometric system to provide security in EMG control HCI applications is also proposed.

This thesis first investigates the novel EMG signal analysis approach which allows detection of a wide range of wrist-hand movements with accuracy exceeding 99% without constraining the accurate sensor placement. This is followed by an investigation of low-complexity real-time EMG signal processing system which maintains state-of-the-art wrist- hand detection performance with 48% lower execution time on an embedded ARM processor and 122% on Arduino ATmega controller. It also investigates the novel biometric systems which allow authentication of a person from different wrist-hand movement EMG signals with accuracy exceeding 99% in real-time on the embedded processor for secured EMG-controlled HCI applications. Moreover, an optimised approach for detecting wrist-hand movements after person identification from its EMG analysis is presented. This approach shows 3% improvement in wrist-hand movement classification accuracy than the traditional baseline systems in real-time on the embedded processor.

Date of Award	Dec 2019
Original language	English
Awarding Institution	Queen's University Belfast
Supervisor	John McAllister (Supervisor) & Jesus Martinez-del-Rincon (Supervisor)