| As driver fatigue driving has become one of the main road traffic accidents, how to monitor the driver state reliably on a real-time basis has become a hotspot of current researches.On the basis of a large number of experiments work, this paper achieves eye location, eye tracking, eye state recognition and fatigue detection. The main research contents as follows:(1) This paper proposes a method of eye detection based on near infrared images. In order to locate eyes quickly and accurately, the active infrared for illumination is adopted to produce bright pupil effect and generate the candidate regions. However, the existence of eyeglasses and other reasons have a negative effect on selection of candidates. Regarding these conditions, the symmetry transform is exploited. Then the SVM algorithm is then used to verify the pupil candidates. Finally, we find accurate pupil center location by means of image gradients.(2) This paper presents an integrated eye tracker by combining Kalman filtering tracking with mean shift tracking. The Kalman filtering is used to track images with bright pupils. When the bright pupils appear weak or disappear, Kalman filtering is used for motion estimation, then the mean shift tracker is activated to take over the tracking.(3) This paper gives a method of eye state recognition based on hybrid protection. After confirming eye region and pupil center location, first of all, the oblique eye images should be adjusted by rotation, then eye outer rectangle is found by using hybrid protection functions in the known eye regions. Finally, the ratio of height and width of the rectangle determines eye state.(4) In the driver fatigue detection module, this paper comes up with a method by combining PERCLOS and blink frequency. Fatigue state could be detected by the blink frequency and the value of PERCLOS, which is the ratio of closed eye image frames and total image frames. At length, test proves that the method of this paper has good effect and high precision. |
PDF Full Text Request