Development of non-contact real-time blink detection system for doze alarm

It is important to detect the "doze" state at an early stage and to warn drivers in order to prevent accidents caused by doze operation. Doze is the state in which the consciousness level falls owing to lack of sleep, fatigue, illness, etc. When the consciousness level falls, several changes appear, i.e., the number of blinks increases and the closed eye time lengthens. In this study, we developed a real-time blink detection system aiming for application in a non-contact doze detection system from the change of blink characteristics. Further, we developed an electroencephalogram-eye blink correlation analysis system in order to investigate the relationship between the consciousness level and blinking. Grayscale pictures were converted into binary images. The area of the iris was measured using eye positional information obtained by the partial eye template matching method. An eye blink was judged when the area of the iris became below a threshold. The present system detected blinks without contact and in real-time, and measured the number of blinks and closed eye time.