Force Myography Signal-Based Hand Gesture Classification for the Implementation of Real-Time Control System to a Prosthetic Hand

Abstract

This study is aimed at exploring the prediction of the various hand gestures based on Force Myography (FMG) signals generated through piezoelectric sensors banded around the forearm for the implementation of a control system in a prosthetic hand. Matlab, Simulink software has been utilized for the analysis and classification. Several classification and recognition models have been considered, and the Tree Decision Learning (TDL) and Support Vector Machine (SVM) have shown high accuracy results. Both of these estimated models generate above ninety five percentage of accuracy in terms of classification. As the classification showed a distinct feature in the signal, a realtime control system based on the threshold value has been implemented in the prosthetic hand. The hand motion has been recorded through Virtual Motion Glove (VMD) to establish dynamic relationship between the FMG data and the different gestures through system identification. The classification of the hand gestures based on FMG signal will provide a useful foundation for future research in the interfacing and utilization of medical devices.

Investigation on the Sampling Frequency and Channel Number for Force Myography Based Hand Gesture Recognition

“…After downsampling, the RMS [ 10 , 11 , 16 ] feature is extracted from the signal using sliding window technology. As an alternative, the MEAN feature is also adopted in this study, which is the same as the mean absolute value (MAV) feature of EMG signal analysis [ 34 , 35 , 36 , 37 ], because all the FMG values are positive.…”

“…Li et al [ 9 ] used 32 FSR sensors to collect signals (100 Hz) for seventeen actions and obtained an accuracy of 99%. Ha et al [ 10 ] simultaneously recorded FMG signal (25 Hz) and hand gesture signals by a virtual motion glove and demonstrated plausible classification accuracy by SVM classification. Xiao et al [ 11 ] applied FMG in virtual reality and obtained an accuracy of 93.4% for classifying six movements, where the FMG signals are sampled at frequency of 50 Hz.…”

“…Pattern recognition technology is usually used to predict hand gestures from FMG signals. As summarized in Table 1 , RMS is one of the most commonly used FMG feature [ 11 , 13 , 16 ], and SVM [ 9 , 10 , 13 , 16 , 17 , 21 ] and LDA [ 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 22 ] are the most commonly used classifiers for FMG based hand gesture classification. Although different features (or systems without feature extraction) and different classifiers are applied in this field, one hardly finds which type of feature or classifier outperforms the others as seen in Table 1 .…”

Biosignal-Based Human–Machine Interfaces for Assistance and Rehabilitation: A Survey

“…The sensor was placed on a forearm muscle, proving to be as effective as the EMG envelope to control a hand prosthesis prototype [155,156]. Ha et al explored the prediction of hand gestures by applying piezoelectric sensors around the forearm to map muscle contraction [157,158]. A piezoelectric sensor converts its mechanical deformation due to the applied force into an electrical signal.…”

A Review of Force Myography Research and Development

“…For instance, Li et al used five flexible piezoelectric film sensors around the thigh to predict four leg movements, and they were able to achieve an accuracy as high as 92% across four participants [56]. Ha et al used three piezoelectric sensors on the flexor carpi radialis, flexor carpi ulnaris, and brachioradialis muscles to predict four upper limb gestures [57]. They were able to achieve an average accuracy of about 80% across three participants.…”