Research, Based On Computer Vision Driver Fatigue / Drowsiness Detection Methods

Driver fatigue/drowsiness is one of the important causes of crashed accidents. Surveys show that 7% accidents are caused by drowsiness/fatigue directly, and 40% of fatal accidents, or 35% of heavy truck and highway accidents. It has been proved that all accidents caused by sleepiness, fatigue and inattention are due to or directly relate to drowsiness. It is reported by Ministry of Communications that both absolute number and proportion of our country's traffic accidents are highest in the world. So we must take some actions for this problem. Since accident doesn't happen as soon as fatigue/drowsiness takes place, real time fatigue detection system can work. Therefore, many countries make great efforts on how to detect drowsiness during driving.At present, systems that purport to measure fatigue fall into one of the two general categories: 1) Simple systems that are currently commercially available, but with uncertain effectiveness in terms of reliability, sensitivity and validity; 2) More complex systems, such as "Drowsy-Driver Detection and Warning System" by NHTSA, that seem potentially effective are undergoing rigorous evaluation and design.In this thesis, we investigate and contrast the principles of current methods and technologies for driver fatigue detection, and analyse the key problems and difficulties of these techniques, then find my own way to detect driver fatigue quickly and effectively. We lay a camera on the panel to shoot the driver's face. Eye features, which are cues for determining driver's state, are extracted from the video of face. We do the above things under two conditions:1) Under natural illumination. Natural solar do no harm and disturb to people, and the system is friendly and simple, so it is worth studying. Skin color is used to locate face. After skin segmentation, accumulative histogram threshold is used to binarize the face image, then the eyes are separated from the skin color perfectly. This method is robust fordifferent intensity. 4-connected component algorithm is used to calculate area of two eyes, avoiding step of separating two eyes, then the open degree of eyes is gotten. Two visual cues -Blink Duration andPERCLOS are used to determine whether there is fatigue taking place.2) Under infrared illumination. Under infrared illumination the system can work the whole day, which can deal with the night driving. Using the eye characteristics at special wavelength, we make a new eye location algorithm, which can improve the eye tracking speed and accuracy. This system can use at any conditions.We have designed energy functions of eye template, using deformable template matching algorithm to extract eye features, including eyelid edge, iris edge, pupil center and so on.