Study On Driver Fatigue Detection Algorithm

Driver fatigue is one of the most important causes of numerous traffic accidents.It is of great social and economic value to develop a driver fatigue monitoring and early warning system which warns the driver when he is fatigue,so that it improves driving safety and reduces even prevented traffic accidents.In this thesis,a non-contact,real-time driver fatigue detection system is proposed based on vision-based method.The system extracts the PERCLOS value of the eye to judge the driver's fatigue state in the way of image processing.Research on illumination compensation of face images,face detection,eye detection and tracking has been carried out in this thesis.The main contributions are as follows:Firstly,the uneven illumination of image is due to the illumination angle,occlusion,and so on during image acquisition process.It can change the original information in a degree of a picture and disturb face and eye detection.A homomorphic filtering method is used in this thesis to enhance the brightness and contrast of the image with well maintaining information of detail and color of skin?Secondly,this thesis adopts the method of combining color region segmentation and Adaboost classifier.Segmentation based on skin color can narrow the region of the face and the face classifier in a high accuracy solves the exact location of face problem that the Gaussian skin color model can not do.This thesis takes full use of the advantages of two kinds of detection methods to improve the speed and accuracy of face detection.Thirdly,During the normal driving process,the motion of driver's head will not appear fast moving or wide-angle deflection and no external objects will be sheltered,a new algorithm for eye detection and tracking based on MOSSE(Minimum Output Sum of Squared Error)algorithm is proposed.The algorithm uses the real-time filter trained in the first frame of image with detected eyes to detect and track the human eye in the next frame and achieve better detection performance.Fourthly,a driver fatigue judgment method is proposed based on the principle of PERCLOS.After locating the rectangular of the human eye,the proposed method can identify the eye state according to pixel feature,which included the length width ratio of rectangle region and the proportion of black pixels to total pixels and black pixel ratio in the 1/2 area of the eye center.According to the idea of fuzzy comprehensive evaluation,these three indexes are combined into a discriminant function by different percentages.Finally,the eye state is determined by comparing the function value with the threshold value.In this thesis,the algorithm is studied based on MATLAB and the experiment is carried out with OPENCV,which achieves the real-time detection rate of 25 frames per second.