The fatigue warning system is very important to guarantee driver′s safe driving. The discrete fatigue estimate model of percentage of eyelid closure over the pupil time bases on image process of eyes is established, and the driver fatigue status monitoring and warning are realized. Eye images acquisition is realized by complementary metal-oxide-semiconductor camera with near-infrared lamps shine. Adaboost algorithm is applied to locate the eyes, and Harris corner detection is used to obtain the information about sight of eyes. Grade warning level is designed by the adjustable warning threshold based on the fatigue estimate model. The experimental results show that the system can provide under 1.5 s reaction time, 4% false alarm rate, and possess good real-time and anti-jamming characteristic under certain conditions.