The Research Of Fatigue Driving Detection Based On Facial Information

With the rapid development of transportation industry, the number of motor vehicles is increasing, and the frequency of traffic accidents is also growing, which has brought huge personal security risks and economic losses to the people. Driver fatigue is a major factor affecting safety driving. According to statistics, about 20 percent of traffic accidents are caused by the driver’s fatigue driving. Therefore, it is especially important to make real-time and accurate fatigue warning for the drivers before the accident. After analyzing a variety of fatigue detection methods domestic and foreign, we study three facial features,the eyes, mouth and nose in this paper, by fusing multiple parameters improves the accuracy of detection of fatigue. In this paper, the main contents are as follows:1. Video image preprocessing. Taking into account the effect of light and noise during driving, after comparing different methods of image enhancement, select median filter and Laplacian sharpening operator to enhance the original image; after comparing the effect of the histogram equalization and the reference white algorithm,we select the improved white reference algorithm to compensate the image.2. Face detection and tracking. First using color characteristics to detect the driver’s skin color area, and then using integral projection method combined with the proportion of face to locate the face. This method can reduce the influence of neck skin. Finally, combined with the face region detected in the first frame, utilizing Camshift algorithm do forecasting and tracking for the face region in subsequent frames. This method consumes less time and enhances the real-time detection.3. Eye detection and doze judgment. First, do integral projection to the upper half of the face binary image, according to the trough projection curve to the detect the eyebrow eye region; then analyze the the curve of horizontal and vertical projection of the eyebrow eye area, and calculate the ratio of the trough width of the two projection curves, combined with the proportion of black pixel in eyebrow eye region to judge the state of eyes; finally, based on PERCLOS criterion and the eye blink frequency to judge whether the driver is dozing off.4. Yawning judgment. First, use canny operator to extract the edges of lips and nostrils in the binary images of the mouth and nose, by calculating the ratio of thedistances from the upper and lower lips to the nostrils, to determine whether the mouth is open; then use the duration of open mouth to judge whether the driver is yawning. This method can reduce the impact generated by the change in distance and enhance the accuracy of the system’s detection.When judging fatigue, first detect whether the driver is dozing off, if yes, then the fatigue warning will be sent to the driver directly; otherwise, go into the next step,detect whether the driver is yawning, if the yawning is detected, the system also will send a fatigue warning. In the experiment, we use video images of 5 experimenter to detect whether they are fatigue, by fusing three characteristic parameters the eyes,mouth and nose, the system improves the accuracy of fatigue judgment, and the effect is good.