Research On Driver Fatigue Test Measurement System Based On Identification Of Human Eye

In 1885, the engineer named Karl Benz from Germany created the first automobile of the world in Mannheim. July 13, 1956, the first automobile named Jiefang which made in China was produced in Changchun, Jilin Province. Experienced development of 50 years, China is moving from the car power to a vehicle power. As the statistics, Chinese auto sales in 2008 was 9,380,500 with the growh of 6.7%. However, with the increasing number of cars, many social problems would follow. Traffic safety, traffic congestion and environmental pollution has become the three major problems of international road transport, and traffic safety is the first place. Human factors, such as drink driving, fatigue driving, is the main reason of traffic accidents. Therefore, in order to reduce the rate of traffic accidents and the number of casualties caused by fatigue driving, and reduce the safety risks caused by human factors, driver fatigue test measurement system was studied in this paper.Through searching a large number of domestic and international data which about fatigue driving, and according to the literature on the measurement methods of driving state, the author predicted the test measurement system of non-contact measurement would be the future direction of development. Therefore, driver fatigue test measurement system based on identification of human eye was mentioned in this paper. It used PERCLOS and blink frequency measurement to evaluate the fatigue of driver. When PERCLOS is greater than 0.4 or blink frequency is less than 7 times / min, the system will determine the driver is fatigue and alarm the diver.This paper discussed the main elements including face contour orientation, eye location, eye tracking and status judge.(1) In the process of face contour orientation, due to influence of video input devices and other factors, color images more or less tended to be warmer or colder. Therefore, the author needed to compensate for color image equalization in order to reduce the influence of light. And according to the characteristics of skin color information in YC_bC_r color space, sub-skin model method was mentioned to detect skin color. The author first established a sub-Gaussian model which is according to Gaussian model is similar to skin color model.,and chose a different skin color model by calculating the brightness value of skin color point, and calculated skin color probability of each pixel,. Then used normalization processing to establish grayscale of skin color probability. And the author used Otsu method to calculate the threshold of grayscale of skin color probability, and used binarization and morphology to segment skin color region.(2) In the process of eye location, the method based on gray level projection and HSI color space was mentioned to locate eyes. The author first used level gray projection to locate eyebrow region in order to narrow the searching scope of eyes, then used the characteristics of the iris pixels in HSI color space to initial positioning the eyes, and used symmetry to rule out candidate regions of single eye. Then the author selected the larger Complexity region to define eye region by calculating the complexity of eye area and eyebrow area, and extracted eye template.(3) In the process of eye tracking, the method based on Kalman filtering and template matching was mentioned to track eyes. The author first initialized eye template, then used the 5 characteristic equations of Kalman filter to forecast the position of eyes in the next frame, and made estimated position as the center, used template matching to locate eyes in the set range, and calculated the maximum of the similarity of matching. If this value was greater than the set threshold, the system was considered that matching was success and made the template update. If this value was less than the set threshold, that matching was failed, the system would re-positioning and searching eyes.(4) In the process of eye status judge, according to the characteristics of iris saturation information in HSI color space, the method based on iris pixel was mentioned to judge eye status. The author first used iris saturation information to establish saturation grayscale of eye tracking and locating pixel in HSI color space, then selected gray level threshold with pixel brightness to judge whether this pixels was iris pixels. Then the author calculated the percentage of the number of iris pixels in the total number of template pixels, if this value was greater than the set threshold, that eyes were open, or eyes were close.Experimental results show that the detection rate was 20 frames / sec and the accuracy of status judge was 84.5% by this method in the video stream, and this approach was satisfied with the system requirements of real-time, accuracy and robustness.