Eyelid Movement-based Driver Fatigue Detection

At present, more traffic accidents have taken place because drivers have been in fatigue. In this thesis driver's fatigue level has been estimated according to his eyes state in order to reduce these traffic accidents, because eyes movement is not only a major feature of driver's fatigue, but also a key of extracting other vision features which express the driver's fatigue level. In this paper, the methods of detecting and tracking eye features of a driver have been studied and improved importantly based on face and eyes of the driver have already been located.In this thesis the following facets work have been finished:Firstly, the eye feature has been extracted on a located eye image. Two steps proposed in this thesis have been used to extract eye features, one of which is iris detection and the other is eye outline extraction. The former is based on Color Segmentation, contrast to the iris detection methods from gray edge images, the accuracy increases consumedly. Based on the iris position extracted, we continue to detect eye outline extraction by using deformable template, contrast to extract iris and eye outline at a time by using it, our method not only can avoid that deformable template collect to the iris district or converge the incorrect region, particularly to a closed eyes, but also can contribute to track and recognize eyes state in following works.Secondly, after extracting eye feature, a novel eye tracking method has been proposed. In order to track the eye accurately when eye is closed, we track iris and eye outline simultaneity. The iris has been tracked from the image sequence by detecting iris half circle using Hough transformation. Because of the known iris radius detected in former sequence, and to a same person (a driver), the change of iris radius is very small, the iris radius can be a known parameter of Hough transformation in following sequences. In addition, the iris center position also can be a reference position in following sequences. So the calculation complication reduces consumedly and the speed quickens considerably. To the eye outline tracking, optical flow is used to estimate the eye key points location. Based on them, we can obtain a new eye outline that can be regarded as theinitial location in next frame, and then, we can modify it using deformable template. This kind of method has avoided appearing inaccurate tracking when the eye outline has a large deformable because of open or close eyes, and improved the accuracy of tracking.Thirdly, after detecting and tracking the eye feature, the next step is to recognize the eyes state. In this paper we put forward a method that can recognize the eyes state accurately from the parameters extracted in former steps. First, we can judge the open eye by iris parameter. If it doesn't occur, then we can judge the eyes state by calculating the parameter of the width between the top and bottom eyelids of the eye outline template. This method has avoided the thing that the iris doesn't appear but the eye is still open. So the accuracy of recognition has improved considerably.Contrast to the old method of driver fatigue detection based on eye movement, that is: after eye location, the whole eye image has been tracked, and then the eye features have been extracted from each tracked frame to recognize whether the eye is open or close, or the eye template established on the correlated eye information has been matched with each eye state template which has been trained, and the final eyes state has been decided on the close match template. In our method, the tracking speed and precision have been improved considerably, because though we add a step to extract eye features in initial frame, but we reduce a step to extract eye features to recognize eyes state in following frames, and we can do it by obtaining the eye feature parameters from each tracked frame.