A blink is one of physiological signals that indicates a consciousness level or a drowsiness. Using a Doppler sensor is one solution to realize non- contact blink detection without cameras. However, it is difficult to detect blinks using a Doppler sensor because of low signal to noise power ratio (SNR) of reflected signal from an eyelid. We previously proposed the blink detection method based on machine learning and show that the method provides high probability of detection of blinks. However, it is necessary for the method to do a prior learning process for each person or environment. In this paper, we propose a high accuracy blink detection method without prior learning process. By applying constant false alarm rate (CFAR) processing to the blink signal detection, the method can be robust against the fluctuation of a body or noise, and the probability of detection can be improved. Moreover, we apply two kinds of classification based on the characteristic of the signal waveform of the blink. Thereby, a subtle body movement that is easy to be wrongly detected as a blink is excluded, and false positive in detection can be reduced. We conducted three experiments to evaluate the detection accuracy. As a result, we show that our proposal achieved high detection accuracy of around 99 %.
|Title of host publication
|2016 IEEE Global Communications Conference, GLOBECOM 2016 - Proceedings
|Institute of Electrical and Electronics Engineers Inc.
|Published - 2017 Feb 2
|59th IEEE Global Communications Conference, GLOBECOM 2016 - Washington, United States
Duration: 2016 Dec 4 → 2016 Dec 8
|59th IEEE Global Communications Conference, GLOBECOM 2016
|16/12/4 → 16/12/8
ASJC Scopus subject areas
- Computational Theory and Mathematics
- Computer Networks and Communications
- Hardware and Architecture
- Safety, Risk, Reliability and Quality