Research On Driver Fatigue Detection Based On Video Analysis And Information Fusion

Driver fatigue is a major cause of traffic accidents. With the development of science and technology, using various detection methods to detect driver's mental state has become possible. Alarm and take appropriate measures before accident occurred have important social significance and economic significance in reducing incidence of traffic accidents, casualties and property losses.In this paper, external symptoms that related to driver fatigue have been detected by using video image analysis technology. The related information is analyzed based on information fusion theory. The grade of driver fatigue is determined according to the analysis results. Main work is as follows:Image light compensation method has been studied. Image local dark caused by uniform illumination will have a serious impact on image processing. To solve this problem light compensation approach is proposed based on combination of image decomposition and Multi-Scale Retinex (MSR) methods. The approach can enhance the brightness of dark areas effectively under the premise of preserving image detail. Simple method is proposed to get high frequency component of image. Color illumination image is reconstructed using scale factor. The color image is balanced using S function.Driver face detection and location has been researched. Driver face area is detected and located using combination method of frame difference and skin-color detection according to the characteristic of driver video. The traditional frame difference method is improved to enhance the anti-jamming capability.Driver eye location approach has been studied. Skin-color segmentation method is proposed for facial image binarization to reduce the eye detection region. Find out the relationship between texture feature of candidate image block and the image block of eye region. Study on the fusion in different directions for eigenvalue of texture feature. The detected eyes are verified used blob detection method. Blob detection is researched in the Difference of Gaussians (DoG) space. Adopt improved Geodesic Active Contour (GAC) model to track the eyes in view of the characteristic of complex shapes and non-rigid deformation of eyes. Using narrow band level set method to reduce the amount of calculation data. Mouth detection approach has been proposed based on characteristic of facial feature distribution and difference of color. Find out standard color region and mouth candidate region based on distribution of facial organs. The difference of H component in HSI space for standard color region and mouth candidate region is analyzed. Map out mouth region by the difference.Correspondence quantitative relationship between eyes, mouth, head and the image information has been analyzed. Adopt ellipse model to calculate the percentage of eyelid open. The variation rule between the projections in image of head features triangle and head movement is studied. The state of the mouth is analyzed by calculating the ratio of width and height of mouth.Structural road detection approach has been studied. A method for smoothing the road image based on neighborhood principal component analysis is proposed. To enhance image contrast presented gray pixel stretch means based on Sin function. Present a lane location method used vertical integral projection for image partition. Present a new curve smoothing method to overcome the projected curve changed easily when smoothing. This way can maintain most of the original shape of the curve. Adjustment method of lane edge points is put forward in order to get more accurate lane edge points. Adopt least square method fitting the edge of lane. Find lane region of interest based on the continuity of the lane video. Achieve accurate lane tracking by using steerable filters.Unstructured road detection approach has been studied. Present ideas of road image resampling and large-scale smoothing for the characteristics of unstructured road. It is analyzed of the impact of image resampling. Make out extreme points and segment region by histogram method. Color image is partitioned by Fuzzy C-Means (FCM) approach to get road as a separate image block. The relationships are discussed between the characteristic of image block and the image of road.Find out a method to judge vehicle departure. The relationship between vehicle location and degree of lane departure is discussed. Using the rate of lane departure to indicate vehicle departure speed is proposed.Present driver fatigue detection method based on information fusion technology. Adopt distributed structure of information fusion according to characteristics of the collected information. A decision is made for driver fatigue used Rough Set (RS) theory in decision-making level. The discreteness and normalization of condition attribute is analyzed according to the RS theory. The quantitative relationship between the value of single sampling and the grade of driver fatigue is studied. Find out minimal decision algorithm to determine the grade of driver fatigue.