Fatigue driving is a major cause of traffic accidents. Making accurate and effective fatigue driving detection has significant implications for road safety. This paper proposes a multi-feature fuzzy inference method for fatigue driving based on facial key points. Firstly, the Supervised Descent Method is introduced into fatigue detection to get more accurate location of facial key points. Secondly, according to the location of the two-dimensional facial key points, three-dimensional head model and camera internal parameters, pitch angle that characterize the head pose is calculated iteratively. Finally, multi-feature fuzzy inference method is adopted to judge the driver state based on the eye-blink, mouth-yawn and head-tilt. Videos from the YawDD dataset and videos taken by ourselves are used to verify the algorithm of fatigue detection. The experimental results show that the average accuracy is 91.6%. The method is transplanted into the Samsung Exynos 4412 embedded development board. After some acceleration strategies applied, the system proposed in this paper meets the real-time requirements.

CICTP