Research On Driver Fatigue Detection Algorithm Based In Video Sequence

The driver will show continuous blinking and yawning under fatigue.This thesis detects the driver's fatigue state by counting and combining the blinking frequency and yawning frequency information.At the same time,aiming at the problem of low accuracy of blink detection caused by wearing glasses,driver's glasses removal algorithm is proposed.The specific work is as follows:1.Analyzing the accuracy and effectiveness of various face detection algorithms,a Gauss distribution model is designed to segment the skin area of the input image,then the driver's face area is accurately located through the template matching algorithm.A face detection algorithm based on skin color segmentation and template matching is proposed for driver fatigue detection.2.Aiming at the problem that the accuracy of fatigue detection algorithm decreases when drivers wear glasses,this thesis proposes an algorithm of glasses removal based on PCA/ICA iteration compensation.This method combines PCA and ICA which can reconstruct the non-glasses face image.Iterative compensation algorithm is used to eliminate the gray boundary between the original glasses area and other areas of the face,so that the synthesized non-glasses face image is closer to the original.3.Compare the advantages of MB-LBP features with traditional LBP features and Haar-like features,this thesis proposes a driver face key point detection algorithm based on MB-LBP and Adaboost.The coordinates of the 24 eyes and mouth key points are accurately located in the driver's face detection result using the trained classifier.4.Combined with the PERCLOS and fuzzy inference system,this thesis divides the driver's fatigue states into three levels which provide a fuzzy inference algorithm to judge the driver's fatigue degree.Using the coordinates of the key points of the eyes and mouth to calculate the blinking and yawning frequencies to determine the driver's fatigue.The experimental results show that the driver fatigue detection algorithm proposed in this thesis has higher accuracy than the traditional driver fatigue detection algorithm and meets the real-time requirement.