Research And Application Of Fatigue Detection Algorithm Based On Human Eye State

The generation of human fatigue is affected by many factors.Working under fatigue condition will not only affect the quality of work,but also cause loss of life and property in many fields.In the field of transportation,the number of traffic accidents caused by fatigue driving accounts for 20% of the total in China,and more than 9000 people are killed in these accidents.Therefore,the real-time detection method of fatigue state has a broad application prospect which can prevent accidents caused by fatigue driving.This thesis mainly studies the fatigue detection method based on human eye state in driving scenarios.The main work of this thesis is as follows:First,a fatigue detection algorithm based on fuzzy comprehensive evaluation is implemented in this thesis.This method uses the Adaboost cascade classifier to detect human faces and segment the eye area,and performs image preprocessing to reduce environmental interference.Image processing is used to accurately locate human eyes and extract human eye feature parameters.This thesis proposes a method for identifying human eye state and fatigue state based on fuzzy comprehensive evaluation.Experiments show that the fuzzy comprehensive evaluation method is systematic and the results are clear.The accuracy of human eye state recognition on the test set is 95.3%,and the accuracy of fatigue detection is 93.8%.Second,a fatigue detection algorithm based on facial feature point location is implemented in this thesis.This algorithm uses a face alignment method based on ERT(Ensemble of Regression Trees)to locate face feature points,and uses the eye aspect ratio(EAR)to determine whether the human eye is open or closed.A feature correction algorithm based on Euler angle of head pose is proposed in this algorithm,experiments prove that it can effectively reduce the impact of changes in camera angle and head posture so as to be able to extract the fatigue features in the eye area more accurately.This method detects fatigue driving and dangerous driving based on the fatigue features in the eye area and abnormal head movements,and the accuracy rate on the test set is 95%.Third,unlike the commonly used fatigue detection algorithm based on face detection,this thesis applies the target detection technology to fatigue detection,and implements a fatigue detection algorithm based on a deep convolutional neural network nemed SSD(Single Shot MultiBox Detector).This thesis builds the SSD network based on the Pytorch framework,builds the image data set needed for training and mark the human eye state and mouth state of each image in the data set.The experiment proves that the trained SSD network has good robustness to head posture and external interference,and can adapt to changes in complex conditions.The precision of human eye state recognition on the test set reaches 99.7%,and the recall rate is above 96%.The algorithm combines the fuzzy comprehensive evaluation thought and the single index judgment method to determine the degree of fatigue,and can effectively recognize the sober,fatigue and severe fatigue state of the human body.The overall accuracy rate on the test set is 96%.All of the above fatigue detection algorithms can effectively detect the driver's fatigue state.The two fatigue detection algorithms based on face detection have faster detection speed,but poor robustness to head posture and face occlusion.The fatigue detection algorithm based on SSD does not rely on face detection,it has good robustness but it's detection speed is a little slower,which can meet the real-time detection requirements of low frame rate cameras.