The Research On Drive Fatigue Detecting Algorithm Based On The Visual

With the improvement of material conditions, the automobile in daily life is becoming increasingly popular. Although the automobile gave people a lot of convenience, but also brought an increasing number of traffic accidents, making the majority of the family and society suffered enormous damage and loss. According to statistics, in all accidents, the number of accidents caused by driver fatigue accounts for a large proportion. Therefore, numerous researchers from different areas and different countries have carried out research on driver fatigue detection, and have made great achievements. Considering fatigue detection should meet the requirements of high accuracy, real-time, robust and comfort at the same time, we researched the algorithm of driver fatigue detection based on vision in this paper, and verified that it is a non-contact, real time, and high accurate algorithm.We comply the principle of big to small and gradually narrowing position in this paper, extract the driver's eye fatigue characteristics by digital image processing, computer vision and pattern recognition techniques. We determine whether the driver is fatigue or not by a method that combining PERCLOS fatigue criteria and blink frequency. In this paper, the main work is done as follows:1.To denoise for image by using a method of median filtering. To do light compensation for image by using a method of combining histogram equalization and the "reference white". By a simple operation for image pre-processing, can make the processing of subsequent steps easier, and improve the accurat and robust of the algorithm.2. By the feature that skin color clustering in YCbCr model, we can extract suspicious facial region in the original image, reduce subsequent detection target area, improve system speed, better real-time.3. Using adaboost algorithm to detect the face region. Train two face classifier by haar features and LBP features respectively. And then,using two face classifiers to test respectively and compare results. So, we chosed the faster, more acurrate and robust face classifier that trained by LBP features.4. Based on the prior knowledge of "three court five", we can coarse positioning of the human eye in the facial region, Then use eye classifier that trained by LBP features for precise positioning. Extract human eye contour in precise eye position by Otsu binary segmentation and morphological operations. After tilt correction, then calculate the aspect ratio of the human eye contour, we can distinguish between opened eye and closed eye by the size of aspect ratio.5. We can determine whether the driver is fatigue or not by using a method that combining PERCLOS-P80 fatigue criteria and blink frequency. This method has advantages of high accuracy, good real-time and robust, can effectively to detect driver fatigue and early warning.Simulation and Realization of fatigue detection algorithm in this paper is in VS2013 development environment by using C++language and computer vision library of OpenCV2.49. By static simulation of a driver driving environment in cab, carried out test in different light, different blocking and different steering angle can verify the fatigue detection algorithm in the static simulated conditions is achieved real-time and high accurate at various stages of the algorithm. Effectively detect fatigue state of the driver, and early warning to prevent accidents caused by driver fatigue.