Research On Fatigue State Decision-making Based On Information Entropy

Fatigue is a subjective feeling of uncomfortable.Under the same objective conditions,people lost the ability to complete the normal activities and work.Fatigue is a physiological state that everyone have.Seriously fatigue reduce efficiency,even people's lives.In order to reduce the harmful of fatigue,this paper proposes a fatigue level method that based on information entropy.In the paper,we utilize some technologies which are image processing and pattern recognition.Firstly,facial detection is carried out from the frame picture.We extracte the eyes and mouth of features,classified these features.Then the POSIT algorithm is used to estimate the posture of head with the key feature points which are obtained from facial detection.The last but not the least,according to these states of each part of the organs,the fatigue decision-making index can be obtained.The weight of the fatigue decision-making index in the video can be calculated by the entropy-weighting method.The Bayesian method is used to judge the driver's fatigue level in a period of time according to the fatigue decision-making index.The details are as follows:1)In order to reduce the influence of illumination on facial detection,self-quotient map is used for image pre-processing.ASM model is trained by MUCT face database and new fatiguerelated face data set.Therefore,we can obtain better registration results when eyes are closed and mouth is opening to yawn.2)Edge detection and Hough transform are used to locate the center of pupil.HOG features are selected to describe the texture of eyes and mouth,because of the advantages and disadvantages of commonly used image texture features.According to the key facial feature points,the eye and mouth regions are acquired.We extract the HOG features and classify the HOG features by SVM,and the mouth and eye states are obtained.In order to increase the using probability,average area of eye and mouth length-width ratio are used to judge the corresponding state.3)We describe the relationships and transformations of four different coordinate systems,and calibrate the camera.By combining the camera's internal parameters,facial features and POSIT algorithm,we estimates the current frame's head pose.At the same time,the head posture is obtained by using the head posture instrument,and the results of the two methods are compared.4)In order to improve the effect of fatigue detection,multiple decision-making indicators are used to characterize fatigue.By analyzing the changes of human states,PERCLOS,blinking frequency,the maximum time of closing eyes and the speed of opening eyes are selected to describe the fatigue states of eyes.At the same time,the mean speed of pupil scanning of pupil(the fatigue states of pupils),the yawning frequency(the fatigue of mouth)and head posture(the fatigue states of head)are added as fatigue decision-making index.5)Since fatigue is a fuzzy and developing state,the concept of entropy is introduced to determine the degree of confusion.It is unequal to the contribution of fatigue that each fatigue decision-making index.On current image the weight of fatigue states decision-making is obtained by using the entropy method of information entropy.The experimental results show that the accuracy of this method in fatigue decision-making is 90%.This method has higher accuracy than the method of only detecting facial features and the method of detecting pupil direction and head features.This is reason that the paper combines more fatigue-related features and introduces entropy to measure the uncertainty of fatigue,which makes the results more accurate.The research results of the paper provide theoretical and technical support for the application of fatigue state decision-making.