Based On The Facial Characteristics Of The Driver Fatigue Detection

A large part of traffic accidents result from the driver fatigue. Therefore, the detection of driver fatigue has received more and more attention. Although much effort has been paid to the research of the detection method of fatigue, two problems are still challenging. First, the driver's specific environment is seriously affected by the illumination. Second, it is difficult to assess fatigue by a single indicator, thus a composite indicator is quite necessary. Therefore, the detection of driver fatigue problem remains unsolved and it still can not be applied in real world recently.The most obvious change of driver fatigue is that the eyes open and close in an abnormal way. Moreover, the face will change at the same time. Therefore, the key of an effective way to detect the driver fatigue is how to improve the driver's face and eyes detection efficiency in different context, and detect the fatigue state in accordance with the laws of their changes.The work and main contribution of this paper are listed as follows:1. A literature review on driver fatigue detection is given. A framework of a driver fatigue method by detecting the eyes and facial changes proposed.2. For color images, this paper proposes a complexion partition approach called "sub-level light compensation + adaptive threshold selection" based on YCbCr color space. A self-adaptive nonlinear transformation compensation scheme is employed, which compensates the illumination according to the different levels of illumination. A color model is established, and a skin likelihood image is provided. Then face segment is accomplished by using an adaptive threshold value approach. The experimental results show that the complexion segmentation method can effectively overcome the usage of fixed threshold segmentation defects.3. For gray-scale images, a geometric characteristics and enhanced classifier cascade method has been introduced. We adopt the "Haar eigenvalue + AdaBoost" classifier method to detect the driver face. For the selection of AdaBoost weak classifier, we propose an AdaBoost express training solutions based on the AdaBoost face detection algorithm. The proposed method conquers the time-consuming problem of the AdaBoost face detection. 4. In order to determine the position and the status of the eyes, an Unscented Kalman filter is introduced to track the eye of the driver in this paper. By employing geometric features and projection methods to locate the eyes, driver fatigue can be determined by judging whether the eyes close for more than 5 frames or not. The experimental results show the effectiveness of the eye detection tracking method in different circumstances.5. The detection of the driver's mouths is used to determine the status of driver's fatigue state. By determining the ratio between the height and width of the mouth and the duration of yawn, we can judge whether the driver is tired or not.The images used in this paper are pictures taken by our own and from the face database. Based on MATLAB7.0 and VC++ 6.0, experiments have been carried out to detect the position and status of face, eyes and mouth. The experimental results show the effectiveness of the proposed fatigue detecting method.