Research On Driver Fatigue Detection Based On Eye State

Fatigue driving is the one of the significant factors in traffic accidents, which is aserious social problem. Driver fatigue detection based on computer vision is anon-vehicle, real-time and non-contact fatigue detection technology, which has becomea hotspot of current research.In this paper, the main research contents are face detection, eye location, eyefeature extraction, principle and realization of the calculation of fatigue’ degree.Firstly, the classical AdaBoost algorithm is produced in detail. which involvesHaar-Like features, weak classifier，cascade AdaBoost classifier, etc. Next, face isdetected based on AdaBoost algorithm. Then, according to the structure of face’ threechambers and five hole, the approximate region of eye is obtained easily, and theregion of driver’s eye is found. The experimental results show that the proposedmethod can improve the speed of eye detection can satisfy the the requirement ofreal-time system.Although the AdaBoost algorithm with a low false positive rate is fast, there is nodistinction between samples of error classification in the procedure of weight updating,which ignores hit rate of positive samples. In this paper, a new method of weightupdating is proposed to improve the hit rate of positive samples, which pays moreattention to positive samples of error classification.After the region of driver’s eyes is obtained, the ellipse fitting algorithm based onleast squares method is used to fit the eye contours, and the eye state is identifiedaccording to parameters of ellipse. Then, the fatigue state could be detected based onPERCLOS and blink frequency.Finally, a driver fatigue detection system based on eye state is designed, and theexperimental results show that the average accuracy of fatigue detection can be up to83%. It has a good performance of fatigue detection.