Person Identification by Deep Learning using Walking Signals Detected by Ultra-Sensitive Electrostatic Induction Technique

摘要

In this paper, we develop a technique to measure walking motion by using ultrahigh sensitivity electrostatic induction technique. As a result of walking motion measurement using this technique, it is obvious that the detected electrostatic induced current waveform has a peak occurring at the timing of foot contact or detachment due to walking motion. Based on these results, a theoretical model in which an induced current is generated is constructed, and the correspondence relationship between walking motion and electrostatic induction current is shown. Furthermore, when comparing the walking signals of each subject, the authors used a scalogram obtained by wavelet transforming the walking signal. Person identification was attempted by learning the subject's scalogram using Chainer which is one of the frameworks for deep learning.