Korean Finger Number Gesture Recognition Based on CNN Using Surface Electromyography Signals
- PDF / 1,077,513 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 106 Downloads / 199 Views
ORIGINAL ARTICLE
Korean Finger Number Gesture Recognition Based on CNN Using Surface Electromyography Signals Jong‑Jun Park1 · Chun‑Ki Kwon1 Received: 4 April 2020 / Revised: 10 July 2020 / Accepted: 15 October 2020 © The Korean Institute of Electrical Engineers 2020
Abstract This study proposes a recognition strategy for Korean finger number gestures based on convolutional neural network (CNN) using surface electromyography (sEMG) signals. A few studies have reported Chinese finger number gesture recognition using sEMG signals. However, finger number gestures vary across different regions, prompting the need to investigate finger number gesture recognition specific to Koreans. To this end, six Korean finger number gestures ranging from zero to five were selected and recognized by CNN using sEMG signals acquired from four pairs of electrodes on forearm muscles. In this study, we investigated the feasibility of CNN in finger number gesture recognition using sEMG time series data. The experimental results show that CNN achieved a 100% recognition rate over six Korean finger number gestures using sEMG time-series data. A comparative analysis of different studies indicates that the proposed approach may be at least comparable to the existing studies selected in this work. It is therefore a more convenient and promising platform for recognition of finger number gestures. Keywords Convolutionary Neural Network · Korean finger number gesture · Surface electromyography signal
1 Introduction The role of surface electromyography (sEMG) signals, which are non-intrusive bio-signals expressing human intentions, has been investigated in the recognition of a wide range of human actions [1–6]. The sEMG signals have been used as a simple on/off switch in the early stage when only a single intention could be defined. As related technologies generate sEMG signals capable of defining multiple human motions, the use of sEMG signals has been extended to complex systems such as navigational control of poweredwheelchair [4]. Further, it has placed visual image data in hand gesture recognition since image acquisition is restricted in area and is very sensitive to lighting conditions [5–7]. For instance, Chen et al. in [5] classified 10 Chinese finger number * Chun‑Ki Kwon [email protected] Jong‑Jun Park [email protected] 1
Department of Medical IT Engineering, Soonchunhyang University, Soonchunhyang‑ro 22‑20, Asan‑si, Chungchungnam‑do, South Korea
gestures using sEMG signals based on classical techniques. The accuracy ranged from 65% to 97.93% depending on the combination of extraction techniques and object classifiers in classical technique, based on a trial and error approach depending on individual experience. To overcome the limitations of manual selection of appropriate combinations in classical technique, researchers have turned their attention to convolutional neural network (CNN) as a powerful image recognition solution. Since CNN learns discriminative features from raw image data and performs automatic classific
Data Loading...
