Driving Fatigue Detection Based On Visual Information

With the continuous increase of car ownership in China,traffic safety has become an increasingly prominent social problem,and fatigue driving has become one of the main causes of traffic accidents.Therefore,This paper studies fatigue driving detection and uses a vision-based method to analyze the driver's eye state to determine whether the driver is fatigued.Firstly,this paper performs a series of preprocessing on the acquired image,such as graying and denoising,then performs face detection on the image.Eyes are detected in the face area.The Haar features and MB-LBP features are used to train the face detector based on Adaboost algorithm,and the detection speed and accuracy rate of the two detectors are compared by experiments.Experiments show that the difference between their accuracy rate is not obvious,but the detector using MB-LBP features has an advantage in training speed and detection speed.Then,this paper uses two improved methods to judge whether the right eye is open or closed.Method 1: In order to improve the accuracy and efficiency of the detection,a rough localization method is used to determine the approximate area of the right eye,and then the eye detector is used to obtain the image of the right eye area.Next,the Otsu algorithm and the region growing method are used to perform segmentation and the like on the right-eye image to accurately extract the right-eye binarized image.In this paper,an improved method for calculating the aspect ratio(EAR)of the human eye is proposed.The number of pixels in each column with a gray value of 255 is counted in the eye binary image,and the maximum value is the height of the eye,which effectively improves the the calculation accuracy.For the same test set,the success rate of the judgment whether the right eye is open or closed is increased from 90.8% to 95%.Method 2: Using the face key point detection algorithm based on the Ensemble of Regression Trees to obtain six key points describing the contour of the right eye,and proposing a method of combining the key point coordinates and image processing to calculate the EAR of right eye.At first,the partial image of the right eye is obtained through the key points,then the image is binarized,segmented,etc.by using the Otsu algorithm and the region growing method.The number of pixels of each column having a gray value of 255 is counted in the binarized image and take the maximum value for the height of the eye.The method uses the key point coordinates to calculate the eye width.It has been proved by experiments that for the same test set,the improved method increases the success rate of the judgment whether the right eye is open or closed from 93.4% to 97.2%.In addition,comparing the two improved methods,both of them have good real-time performance,but the latter has higher accuracy and strong robustness under different illumination conditions.Finally,this paper combines the blink frequency method with the PERCOLS criterion to determine the fatigue state.The above two improved methods are compared by experiments.The experiments show that the two improved methods have improved the detection accuracy,but the first method is easy to determine the closed eye as a open eye under the side light condition,resulting in a relatively poor detection results,and the method based on the combination of key points and image processing has better accuracy and robustness in backlight,sidelight,and night dim conditions.