Exploration of Force Myography and surface Electromyography in hand gesture classification. (March 2017)

• Record Type: Journal Article
• Authors: Jiang, Xianta; Merhi, Lukas-Karim; Xiao, Zhen Gang; Menon, Carlo
• Is Part Of: Medical engineering & physics. Volume 41(2017)
• Page Start: 63
• Page End: 73
• Publication Date: 2017-03
• Subjects: Force Myography -- Electromyography -- Wearable sensors -- Hand gesture recognition -- Machine learning; Biomedical engineering -- Periodicals; Biomedical Engineering -- Periodicals; Physics -- Periodicals; Génie biomédical -- Périodiques; Biomedical engineering; Electronic journals; Periodicals; 610.28
• DOI: 10.1016/j.medengphy.2017.01.015
• Languages: English
• ISSNs: 1350-4533

Abstract

Abstract : The FSR band (FMG) achieves as good results as high quality sEMG technology. FMG on the forearm achieves slightly higher accuracy than on the wrist. sEMG on the wrist yields comparable performance to sEMG at the forearm. Grasp gestures have lower classification accuracy than sign language and finger/hand movement gestures by both FMG and EMG. Abstract: Whereas pressure sensors increasingly have received attention as a non-invasive interface for hand gesture recognition, their performance has not been comprehensively evaluated. This work examined the performance of hand gesture classification using Force Myography (FMG) and surface Electromyography (sEMG) technologies by performing 3 sets of 48 hand gestures using a prototyped FMG band and an array of commercial sEMG sensors worn both on the wrist and forearm simultaneously. The results show that the FMG band achieved classification accuracies as good as the high quality, commercially available, sEMG system on both wrist and forearm positions; specifically, by only using 8 Force Sensitive Resisters (FSRs), the FMG band achieved accuracies of 91.2% and 83.5% in classifying the 48 hand gestures in cross-validation and cross-trial evaluations, which were higher than those of sEMG (84.6% and 79.1%). By using all 16 FSRs on the band, our device achieved high accuracies of 96.7% and 89.4% in cross-validation and cross-trial evaluations.