Driver Fatigue Detection Based On Several Fatigue Parameters

According to the statistics from the Department of Transportation, the number of deaths in road traffic accidents is the highest for several years in our country, and driver's fatigue is one of the major factors that cause the traffic accidents. For accident and social property prevention, it is essential to develop method to actively monitor the driver's level of fatigue and alert the driver of any insecure driving conditions.In the thesis, by studying the research status of fatigue detection in domestic and abroad and the characteristic of the real-time monitoring system, the total scheme based on the existing literature is as follows.First, driver's face detection applies the human face skin color model and the characteristic of the human face. The median filtering is used for image pre-processing to eliminate the noise. Then we used the projection functions to locate the face.Second, after obtaining the face window, a novel projection, i.e. hybrid projection function(HPF) is proposed to find the center of pupil in the above half face in order to improve the disadvantage of the ordinary projection function. And it is worked out to have a thick shin to the stochastic noise.To find out the lip region, canny operator is used for edge detection in the below half face . Considering that lighting and other factors can cause the lip and skin color changing, a improved method is proposed to detect the lip corners which select different thresholds of the lip color under different circumstance.Then, the pupil and lip corners tracking are done by different methods. Because of the stability of the lip corners position, a bigger mouth rectangle in the center of the original mouth rectangle is given and then find out lip corners based on it. The pupil tracking is finished by the unscented Kalman filter(UKF) which is a new approach for filtering nonlinear systems. It can restrain noise effectively and shows a good tracking result under the environment with big flexibility.In the end, three fatigue parameters are worked out based on pupil localization, such as PERCLOS, Eye Closure Time(ECT), Eye Blink Frequency(EBF). At the same time, the driver's nod frequency(NodFreq) and head orientation are got based on the corners of lip. Then the instability of single fatigue parameter will be improved and the security during driving will be better ensured.The experiment results show speediness and validity of our method for driver fatigue detection. The research which have been done will be good theory foundation to driver fatigue detection system.