Detection Of Eye State Parameters From Image And Its Use In Drowsiness Monitoring

Driver's fatigue is one of the main causes of traffic accidents. In order to take the initiative to eliminate the factors that lead to traffic accidents and to minimize loss of life and economic, the most thorough way is to use pre-prevention and technical means to monitor and limit improper driving behavior such as the fatigue of driver.This paper describes the research background and the significance of driver fatigue detection. We also compare and analyze the development and research status of driver fatigue detection. A non-invasive approach for monitoring driver's state based on visual information is a better and relatively easy to implement choice. The eye state is the most direct and effective response to the physical characteristics between features used to monitor driver's state using visual methods. The main works of eye detection, eye feature extraction and eye tracking presented in this paper are as follows:Eye detection and eye feature extraction under visible light condition:After face detection using Adaboost face classifier, we detection eyes in region corresponds to the first rectangular feature of AdaBoost face classifier using AdaBoost eye classifier. Taking advantage of different gray distributions between open eyes images and closed eye images, we proposed a statistic model based on linear predictor error distribution of wavelet coefficients to extract eye state feature. Experiment results with the classifier build on these eye state features and support machine vector (SVM) demonstrated that our method is effective. In the eye parameter extraction, we calculate the pupil center and radius based on the characteristic that the pixel values of pupil are located in the valley of eye image. The RANSAC(random sample consensus) method has been applied successfully to the process of extracting eyelid parabolic parameter which makes use the eye edge information.Eye detection and tracking under infrared light condition:In order to overcome the impact of light on image collection under visual light condition, we adopt active infrared light source to obtain the images. Placing the single-ring infrared light source close to the camera optical axis, the bright pupil image can be obtained due to principle of infrared imaging. Calculating the mathematical morphology open operation on the enhancement image of bright pupil image, we can get simulated dark pupil image. The candidate eye region can be selected on the differential image between the bright pupil image and dark pupil image. In accordance with a priori knowledge of the location of the eyes, the final decision whether there is eye in the image and eye position can be made by removing of redundant candidate region. For shortening the time of eye detection in video sequence, we track eyes and detect the presence of eyes and eye position in the search region by using Kalman filter.