Research And Early-warning Of Driver Fatigue Detection Based On Vision

As people's quality of life continues to rise,road traffic is constantly developing and the number of cars is increasing,which is accompanied by the frequent occurrence of traffic accidents.Fatigue driving is one of the most important causes to result in traffic accidents.Therefore,the design of driver fatigue detection system has become the focus of current research,and a driver fatigue detection system that is easy to use,accurate in detection and fast in operation is of great significance to the improvement of frequent traffic accidents.This thesis presents a set of fatigue detection methods using image processing techniques.Through the use of the image processing,the face detection and the feature points extraction techniques,we can analyze the facial features of the driver,and then determine the fatigue state.The fatigue detection system in this thesis is composed of the image acquisition module,the image preprocessing module,the face and face feature detection module and the face feature status analysis module.Among them,a detailed study is made on the face and face feature detection module and the face feature analysis module.Firstly,the driver's head image is collected by the camera in real-time,and the techniques such as the homomorphic filtering and the median filtering are used for light compensation and noise elimination.Moreover,Feature extraction method combining rotation-invariant LBP operator and MB_LBP operator is proposed,which improves the defect of information redundancy of the traditional LBP feature,and the description of the face contour is clearer,then the face classifier is trained which is used to detect the driver's head area in real time.Experimental results show that the improved face detection algorithm has higher accuracy and better stability than traditional face detection algorithms.Secondly,this thesis proposes a method based on improved random forest cascading regression to detect the face feature points,By dividing the facial feature points into regions and performing shape regression on each region separately,the human face shape is finally obtained.This method improves the defect of regulation imbalance of the global shape regression.Then,this method is used to locate feature points of driver's face image,and obtain the driver's eyes and the mouth area through ellipse fitting algorithm.Subsequently,a fatigue judging criterion combining the PERCLOS value,the blinking frequency and the yawning frequency is proposed,which improves the accuracy of the judging compared with the judging method based on only the eye information.Finally,by using the computer as the fatigue detection system simulation running hardware platform and the CMOS camera to collect images,and we do the detection of fatigue experiment in the family car.The experimental results show that the driver fatigue detection system in this thesis has high accuracy and stability,and can meet the real-time requirements very well.