| With the increase of vehicle holdings and the acceleration of living rhythm,traffic accidents due to the driver fatigue were frequently happened.We should pay attention to the serious consequences caused by drowsy driving as well as drunk driving.It is a great hidden trouble of traffic safety and an important issue that threatens the safety of people’s life and property.Therefore,it is necessary and important to research on a reliable and effective technology of fatigue driving detection.The fatigue detection system can make early warning and deceleration through forecast and judgment of driver’s drowsiness in advance to avoid traffic accidents.This paper has summarized the advantages and disadvantages of the different approaches of fatigue driving detection in domestic and overseas.Considering the situation of wearing glasses,the paper presents a comprehensive method based on eyes,mouth and the track of head to identify the fatigue state.The main contributions are as follows:1.Video image pre-processing.The region of driver’s face will be affected by different illumination when driving.And the noise and blurring may generate from the process of acquisition and transmission.So this article makes denoising filter and illumination equilibrium for image in advance to insure the accuracy of face detection.2.After the comprehensive analysis and comparison of different method on face detection,the paper adopts face detection technology which based on Adaboost algorithm to locate the driver’s facial area.It utilizes Haar-Like features and integral image to train the facial classifier iteratively.Compared with the method based on skin color or template,this Adaboost classifier can detect with high efficiency and accuracy.We use the particle filter to track the region of moving target in real time.3.Eyes detection and state recognization.First,the candidate area of the eye is roughly divided by the geometric distribution rule of the facial organ.Second,an adaptive binarization method called OTSU is used.According to vertical integralprojection and the number of connected domain,it can judge whether the driver wear glasses or not.Finally,it has two ways to identify the state of eyes(open/closed).It chooses the method of histogram’s local statistical features to recognize eyes state when driver wearing glasses.Getting eyes region with eyes’ classifier based on Adaboost algorithm when not wearing glasses.Then,this paper binarize the eyes’ area,extract the largest rectangle from one eye and calculate the approximate angle of eye.Judge the state of eye whether open or not by threshold value.4.Mouth detection and state recognization.Mouth has special color and brightness information compared with the other areas of face.In order to simplify the calculation and increase the detection rate,we choose a candidate area of mouth according to the geometric distribution rules of facial organs.Extracting mouth edge with Sobel edge detection.Computing roundness e to identify the state of mouth.This method has rotation invariance.5.The detection of head motion locus.The head will nod when people in the situation of drowsy.This article utilizes the center position of eyes and mouth which are located in previous section to identify the head status.6.Fatigue judging.In this section,the duration time,blink frequency,PERCLOS parameter,yawning duration and times,nod frequency had been defined and taken to detect drowsiness and fatigue synthetically.According to the different level of fatigue on facial expression,this paper builds a fatigue detection flow chart.The system will warn and decelerate the vehicle automatically when detecting fatigue.We took four groups of videos which simulate driving situation to testing.The results show that the fatigue detection algorithm is more efficient and accurate,and satisfies the requirement of real-time system.. |
