Research Of Driver's Fatigue Driving Detection Algorithm

In recent years, with the rapid development of world's economy, the number of motor vehicle is increasing rapidly. As a result, traffic accidents have caused sufficient attention. Studies have shown that when a driver is tired while driving, his situation judgment as well as the control of a vehicle will have a certain degree of decline. Traffic accidents can be happened easily in this case. According to the information displayed, half of the traffic accidents are caused by the driver's fatigue driving every year. Thus it can be seen that drivers'fatigue driving is a big threat to the road safety. In this paper, we use the applied mathematics knowledge and computer vision theory, and establish a mathematical model to detect the driver's fatigue status.The core content of this paper is:important eye feature points'detection; calculate three parameters of eyes that can weigh the driver's fatigue status; establish a model which is based on multiple information fusion.In this article, we need to locate the eye feature points are pupil, internal and external corners of an eye, Purkinje. First, we use the USM sharpening process. To some extent, it will narrow the area of corner detection. Then, using the Gabor filters which are based on the texture information of an image, and we implement the precise positioning of the corners in the area of coarse positioning. Counting the Purkinje's regional characteristics, and obtain a real region of interest based on the iterative filter. We use a method which is based on the least square circle fitting to get the coordinate of the center. Based on this, we begin to calculate three fatigue characteristic parameters of eyes, namely blinking frequency, eyes' closed duration and the angle of sight deviation. Finally, using the support vector machine (SVM) model to integrate the above three parameters, and decide the driver's drowsiness state.In the experiment, we first design a series of experiments to locate the important eyes' feature points of this article, and then calculate three kinds of fatigue characteristic parameters. The final test illustrates the superiority of the method presented in this paper. In the past, we only use a simple fatigue characteristic parameter of blinking frequency or PERCLOS, and the recognition rate is 92% or so. In this paper, the method of information fusion which is based on SVM model increases the recognition rate to 97%. It shows that the method of multiple information fusion in this study improves the veracity and robustness of the fatigue driving detection.