Research And Implementation Of Drivers’ Fatigue Detection Based On Eye Feature

With the growing number of automobiles, the traffic problems have been serious in our nation. Traffic accidents caused by drivers’fatigue driving were more and more. Fatigue driving has become one of the main factors of traffic accidents, In order to guard against occurrence of traffic accidents, it was practical to research and implantation a method used to detect drivers’fatigue.This thesis systemically studied and analyzed the present situation of the fatigue driving both at home and abroad, finally, the driver fatigue detection was implemented based on Adaboost detect algorithm and template matching tracking algorithm. In the process of face detecting, the faces were detected by the face Adaboost cascade classifier in the preprocessing image, and the driver’s face was found after judging all the faces. In the process of eye detecting, the eyes were detected by the eye Adaboost cascade classifier in the face region, and the real eyes were found through judgment of the adaptive threshold. In the process of eye tracking, eye regions were predicted through the displacement of the two frames before, and then template matching was used in these regions. However, in the practical application of traditional template matching algorithm, the eye blinked frequently and the matching inaccuracy accumulated continuously, this gave a result that eye tracking appeared mistakes, and even failure. Facing with such problem, the improved method was proposed in this thesis. The contour of each eye was extracted after template matching, and then eye template was updated. After accurate tracking, the height and width of each eye were obtained, the TOOE was calculated, and then the eye feature was judged. Finally, in order to ensure the accuracy, it was judged that whether detecting or tracking in the next frame.This thesis used VS2008development environment and C#programming language on PC, then implemented this algorithm based on Open CV. Test has been done on real-time and accuracy under different pixel-level, light, movement speed, rotation angles and fringe interference. The result indicates that on the usual sunlight and the640x480pixel-level video, the face detection algorithm, the eye detection algorithm and the eye tracking algorithm introduced by this thesis can be precisely implemented on time in every process.