Study On Driver Fatigue Detection Based On Eye Movement Cues In Rail Transit

In recent years,high-speed railway is rapidly growing in the world,especially in China.High-speed railway needs a higher safety level than other public transports because of more passengers and higher speed.It not only needs higher standards of hardware facilities,but also meets stricter demands to train drivers.It is well known that train drivers play a key role on the rail operation and their physiological and psychological states are important factors to ensure driving safety.Sleepiness and fatigue at work occur frequently in rail transportation due to longer operation time,irregular shift schedule,high psychological and cognitive requirements and so on.At present,it is mainly through improving the management system and the work plan to ease the train driver's fatigue.However,this method can not monitor the driver's mental state in real time.Considering the particularity of rail driving,we designed a fatigue detection system based on eye behaviors.We focus on the key issue in designing software data-acquisition platform,extracting and analyzing characteristics parameters,analyzing individual difference and developing driver's fatigue detection model.The specific research contents are as follows:(1)Designing train driving simulation tests of sober driving and fatigue driving.Eye movement data were collected by a remote eye tracker.KSS,sleep record and detection distance were recorded synchronously.A t-test was applied to evaluate the difference in two driving tests.The designed sober and fatigue driving tasks were proved to be effective.(2)Extracting and analyzing eye movement features.We extracted pupil diameter,blink time and fixation percentage from raw eye data and proved that these three eye movement features can effectively distinguish fatigue from vigilance.Meanwhile,we revealed obvious individual difference of train drivers.(3)Proposing driver-specific classification method.By comparative analysis in BP neural network,BP neural network optimized by genetic algorithm and SVM model,we proposed our driver-specific classification method with multi-parameter combination based on SVM model.The mean accuracy for the driver-specific SVM models is 90.10%,the mean sensitivity is 92.09% and the mean specificity is 86.16%,which demonstrates that pupil diameter,blink time and fixation percentage can be used to detect driver's fatigue state reliably.These quantitative results can be used as a preliminary study for designing human-train interface of high-speed railway to prevent drivers suffering from fatigue.