Driver Drowsiness Detection In Complicated Conditions Based On Computer Vision

Driver drowsiness is one of the leading causes of traffic accidents. Research on drowsiness detection and warning technics is of great significance to improve driving safety. Computer vision technology has been proved to be the most promising technology(high real-time performance, non-intrusiveness, and cognitive consistency) in detecting driver drowsiness through the analysis of driver facial expression. However, there are still many challenges posed by driver’s individual difference, random illumination and unapparent fatigue characteristics when a driver becomes drowsy. This paper focus on the key issues in eye region locating, eye motion extraction, and drowsiness feature space modeling. Real on-road exprements were conducted to testify the accuracy and robustness of the proposed methods.Infrared illumination and infrared filter were adopted to detect drowsiness from the faint facial image at night, and to solve the visibility of eye images when wearing sunglasses. According to a thorough analysis on the influence of facial orientation, occlusion and expression in eye location, frontal face detection combined with a generic 3D face model was used to model the 3D face and to get the head pose of the driver. The eye position of 3D face model and head pose were used to locate the eye position in 2D facial image, overcoming the challenges caused by body movements and enhancing the robustness to complex illumination.The eye position in 2D facial image determined by 3D face model and head pose was used to guide the local constrained local model(CLM) algorithm to locate the feature points in the eye area. Facial image texture normalization was proposed to test the effectiveness of feature points. When detecting failed, tracking strategy was employed to improve the robustness of feature point location. In order to overcome the difference of iris images in different light conditions, iris physiological structures were taken to get the center location of iris. Based on the location of feature point and iris center, parameterized template was used to locate the upper and lower eyelids, achieving the aim of driver eye movement extraction.According to the characteristics of eye movements, drowsy features associated with eyelid opening extent, eyes closed speed and iris motion were extracted respectively. The features in drowsy and awake state were compared to obtain the variation characteristics, overcoming the influence of individual difference. Moreover, the effectiveness of drowsy features in detecting drowsiness was analysed. Bayesian network classifier was constructed to reveal the relationships between the drowsy features, and to discriminate driver drowsiness. Finally, different feature combination was exploried to improve the robustness of drowsiness detection.