The Research Of Eye Movement Index For Detecting Air Traffic Controllers' Fatigue

In recent years,the phenomenon of air traffic controllers' fatigue on their posts has become increasingly prominent.Fatigue will not only weaken the normal working ability of controllers,but also slow down their reaction and decrease their cognitive ability.It is also an important factor that induces controllers to forget and make mistakes,causing unsafe incidents even accidents in Civil Aviation.How to detect the control fatigue and timely warning by technical means is a key and urgent problem to reduce the fatigue error of the controller and ensure the safety of the air traffic control(ATC).The current detection means of air traffic controllers' fatigue mainly includes subjective scale,physiological and biochemical indexes,Psychomotor Vigilance Test(PVT),eye tracking technology and so on.Among them,the eye-tracking technology not only has the objectivity and non-invasiveness that the subjective scale does not have,but also has better applicability than physiological and biochemical indexes,while in the field of vehicle driving fatigue has successful applications for reference.It is direction worth to study for developing real-time fatigue detection and early warning technology.However,at present,there is no systematic study of the performance of many eye-tracking indexes on fatigue detection.In view of this,we recruited 20 students major in ATC to participate in the simulated tower control experiments,and collected eye movement data and subjective fatigue degree(Karolinska Sleepiness Scale,KSS)of subjects from conscious into fatigue state.Through the significant difference before and after fatigue,the correlation analysis with fatigue and Receiver Operating Characteristic(ROC),the performance of each eye movement index detecting controllers' fatigue was investigated.Then we use the Gauss kernel support vector machine(SVM)algorithm and BP model(back propagation)to combine eye movement indexes to establish a fatigue detection model.The conclusion of this study is as follows:1.The fixation point,saccadic velocity,PERCLOS and CFF four indexes not only meet the sober and fatigue data differencing significantly,but also have strong correlation with fatigue and good detection performance of fatigue,which can be used as useful indexes for detecting controllers' fatigue.2.Using the Gauss kernel SVM model and BP model(back propagation)to integrate fixation point,saccadic velocity,PERCLOS and CFF to establish a regulatory fatigue detection model,the results show that the recognition accuracy rate of these two methods is 95.68% and 91.67% respectively.In the future,we can automatically determine whether controllers are awake or fatigued according to the current eye movement parameters and provide timely warning for controllers who enter fatigue state.