Study Of Driver’s Eye State Recognition Algorithm Under Complex Illumination

Fatigue driving, which has aroused widespread concern of society, is nowadays one of themain causes that lead to the traffic accident problem. The non-contact methods of fatiguedetection are effective means to prevent fatigue driving. Among the existing methods, themethod based on PERCLOS (percentage of eyelid closure) index is the mostly recognized onein academia. The key of this method is how to correctly identify the driver’s eyes opening andclosure state. At present, the driver eyes state recognition algorithms and models are notmature. They are difficult to adapt to the complex illumination and other terrible conditions inthe real environment so that it is necessary to carry out in-depth study.This paper analysis the existing eye state features and recognition methods, aims to findeye state features with high robustness to illumination and head rotation, as well asrecognition algorithms which perform good classification to these features, in order to bringusers well experience. The researches of this paper are all conducted on the foundation ofinfrared ray (IR) illumination for the meet of all-day detection need.The main work of this paper is as follows:(1) First, this paper classifies the existing driver’s eye state recognition algorithm, carries outa detailed analysis and comparison of the algorithm under the visible and infrared lightconditions.(2) Second, to adapt to complex illumination of real driving environment, this paper finds agradient histogram feature (HOG feature), that is not sensitive to illumination variation.Combining with SVM mathematical classification model, this paper prevents an eye staterecognition method based on HOG feature.(3) Third, because the geometry of eyes changes greatly when driver’s heads rotate andglasses glisten, this paper presents a novel eye state recognition method based ongray-scale morphology. This method introduces the fuzzy logic theory, divides the eyestate into three states: opening, closure, and unrecognized. It has a high robustness both toillumination and deformation. Besides, it can still work well under other complexconditions.(4) Finally, this paper simulates the proposed algorithm model, and makes a comparison with common eye state recognition methods. The experimental results show that the methodbased on HOG feature has a high robustness to illumination, and the method based ongray-scale morphology is not sensitive both to illumination and eye deformation. In thetrusted image set, the average recognition accuracy of eye state can reach above90%,which is higher than the current common eye state recognition methods. The proposedmethod in this paper can adapt to the need of complex driving environment.