Abstract: A method for detection of drowsiness using visual information of human facial expression was studied and tested. Facial expression features in the state of drowsiness were analyzed based on the characteristics of facial muscle movement and empirical knowledge, then 11 features were characterized. Fatigue driving experiments were conducted using a driving simulator. The features were obtained by the detection of facial markings. Four drowsy expression indexes with significant differences between different drowsy levels have been found, and a discriminating model for drowsiness prediction based on the four indexes has been established, which are eyes open width, eyebrow droop level, corners of the mouth droop level and mouth bending. The results show that the proposed method in this paper could reach an overall correct detection rate of 93%.

Key words: drowsy driving, facial expression detection, expression features, drowsiness detection