Blink detection using Adaboost and contour circle for fatigue recognition

Abstract

Aiming at the problem of traffic accidents, an Adaboost and Contour Circle (ACC) algorithm is developed based on a traditional Adaboost method and the proposed contour circle (CC) for recognizing whether eyes are in open state or closed state. First, Adaboost method is used to detect human faces and eye regions. Second, the pixels of the pupil region are removed by the given grid method. Third, the least squares method is utilized to fit the CC of the upper eyelid. Fourth, the center and radius of the CC are extracted as the feature vector. Finally, the eyes state is recognized according to the defined threshold. It is experimentally proved that the vertical coordinate of the CC is the best feature which can classified whether the eyes are in open state or closed state by the linear decision surface. Besides, the feature vector can classify the eyes states by nonlinear decision surfaces. The correct ratio of the blink detection is 96.6%, and the fatigue blink recognition accuracy is 91.5%.