Driver Inattention Monitoring System for Intelligent Vehicles:A Review Yanchao Dong,Zhencheng Hu,Member,IEEE,Keiichi Uchimura,and Nobuki Murayama
Abstract—In this paper,we review the state-of-the-art technolo-gies for driver inattention monitoring,which can be classified into the following two main categories:1)distraction and2)fatigue. Driver inattention is a major factor in most traffic accidents.Re-arch and development has actively been carried out for decades, with the goal of precily determining the drivers’state of mind.In this paper,we summarize the approaches by dividing them into the followingfive different types of measures:1)subjective report measures;2)driver biological measures;3)driver physical mea-sures;4)driving performance measures;and5)hybrid measures. Among the approaches,subjective report measures and driver biological measures are not suitable under real driving conditions but could rve as some rough ground-truth indicators.The hybrid measures are believed to give more reliable solutions compared with single driver physical measures or driving performance mea-sures,becau the hybrid measures minimize the number of fal alarms and maintain a high recognition rate,which promote the acceptance of the system.We also discuss some nonlinear modeling techniques commonly ud in the literature.
Index Terms—Distraction,driver inattention,driver monitor-ing,fatigue.
I.I NTRODUCTION
D RIVER inattention is a major factor in highway crashes.
The National Highway Traffic Safety Administration (NHTSA)estimates that approximately25%of police-reported crashes involve some form of driver inattention—the driver is distracted,asleep or fatigued,or otherwi“lost in thought”[1]. One common definition of driver inattention is given in[2]:“Driver inattention reprents diminished attention to activities that are critical for safe driving in the abnce of a competing activity.”
A study by the American Automobile Association Founda-tion for Traffic Safety(AAA FTS)utilized the followingfive categories for the driver attention status[3]:
咳嗽不能吃的东西1)attentive;布鲁斯口琴音阶图
2)distracted;
3)looked but did not e;
4)sleepy;
5)unknown.
Manuscript received December24,2009;revid September5,2010; accepted November3,2010.Date of publication December13,2010;date of current version June6,2011.The Associate Editor for this paper was L.Li. The authors are with the Graduate School of Science and Technology,Kumamoto University,Kumamoto860-8555,Japan(e-mail: dong@navi.cs.kumamoto-u.ac.jp;hu@cs.kumamoto-u.ac.jp;uchimura@cs. kumamoto-u.ac.jp;murayama@cs.kumamoto-u.ac.jp).
Color versions of one or more of thefigures in this paper are available online at ieeexplore.ieee.
Digital Object Identifier10.1109/TITS.2010.2092770
The category“looked but did not e”can be considered a kind of cognitive distraction,and the word“sleepy”could be replaced by the more comprehensive word“fatigued.”In this paper,we propo the following two categories for inattention: 1)distraction and2)fatigue.
The caus of driver distraction are diver and po large risk factors—more than half of the crashes that involve inat-tention were caud by driver distraction[1],[2].After an intensive study on t
he various definitions of driver distraction appeared in the literature,a more general definition is propod in[2]:“Driver distraction is a diversion of attention away from activities critical for safe driving toward a competing activity.”Thirteen types of potentially distracting activities are listed in [3]:eating or drinking,outside person,object or event,talking or listening on a cellular phone,dialing a cellular phone,using in-vehicle-technologies,and so on.Becau the distracting activities take many forms,the NHTSA classifies distractions into the following four categories from the viewpoint of the driver’s functionality[1]:
1)visual ,looking away from the roadway);
2)cognitive ,being lost in thought);
3)auditory ,responding to a ringing cell
phone);
4)biomechanical ,manually adjusting the
radio volume).
Two more categories are added in[2]:1)olfactory distraction and2)gustatory distraction.Several distracting activities can involve more than one of the ,talking to a phone while driving creates a biomechanical,auditory,and cognitive distraction).
The phenomenon of fatigue is different from that of distrac-tion.The term fatigue refers to a combination of symptoms such as impaired performance and a subjective feeling of drowsiness [4].Even with the intensive rearch that has been performed, the term fatigue still does not have a universally accepted defi-nition[5].Thus,it is difficult to determine the level of fatigue-related accidents.However,studies show that25%–30%of driving accidents are fatigue related[6].In their definition,the European Transport Safety Council(ETSC)states that fatigue “concerns the inability or disinclination to continue an activity, generally becau the activity has been going on for too long”
[7].From the viewpoint of individual organ functionality,there are different kinds of fatigue,such as the following cas:
古诗鸟白居易1)local physical ,in a skeletal or ocular muscle);
2)general physical fatigue(following heavy manual labor);
3)central nervous fatigue(sleepiness);
4)mental fatigue(not having the energy to do anything).
1524-9050/$26.00©2010IEEE
Fig.1.Information processing and attention[10].
Central nervous fatigue and mental fatigue are the most dangerous types for driving,becau the cas will eventually lead to sleepiness,increasing the probability of an accident. The ETSC defines four levels of sleepiness bad on behav-ioral terms as follows[7]:
1)completely awake;
2)moderate sleepiness;
3)vere sleepiness;
4)sleep.
In an attempt to avoid having an accident,most sleepy drivers will try tofight against sleep with different durations and quences of the physiological events that precede the ont of sleep[8].When a driver becomes fatigued and begins to fall asleep,the following symptoms can be obrved:
1)repeated yawning;
2)confusion and thinking ems foggy;
3)feeling depresd and irritable;
4)slower reaction and respons;
5)daydreaming;
6)difficulty keeping eyes open and burning nsation in the
eyes;
7)lazy steering;
8)difficulty maintaining concentration;
9)swaying of head or body from nodding off;
10)vehicle wandering from the road or into another lane;
11)nodding off at the wheel;
12)breathing becoming shallow;
13)heart races.
Different individuals show different symptoms to varying degrees.Thus,there is no concrete method of measuring the level of fatigue.The ETSC study[7]showed that the level of fatigue or sleepiness(sleepiness is the outside the exhibition of fatigue)is a function of the amount of activity in relation to the brain’s physiological waking capacity.Several factors can influence this physiological waking capacity and,hence, lower the fatigue threshold[4],[5],[7],[9],such as disturbed sleep,the low point in the circadian rhythm,and hard work prior to driving.The factors are independent of the activity being undertaken but result in the fatigue effect of that activity appearing more quickly.Thus,fatigue cannot be en simply as a function of the duration of time engaged in driving.
Driving is a process that involves situation awareness of the environment,decision making,and the performance of actions, as shown in Fig.1[10].In this process,the most complicated stage is the situation awareness.In[10],a three-level situation awareness model is defined as“the perception of the elements in the environment within a volume of time and space,the com-prehension of their meaning,and the projection of their status in the near future.”The deployment of attention in the perception process acts to prent certain constraints on a person’s ability to accurately perceive multiple items in parallel and is a major limitation on situation awareness.Direct attention is needed n
ot only to perceive and process the available cues but in the later stages of decision making and respon execution as well.In a complex and dynamic driving environment,attention demands result from information overload,complex decision making, and the performance of multiple tasks.Thus,monitoring the attention status is vital for maintaining safe driving.
The purpo of the Driver Inattention Monitoring System (DIMS)is to monitor the attention status of the driver.If driver inattention is detected,different countermeasures should
be taken to maintain driving safety,depending on the types and levels of inattention.DIMS has been an active rearch field for decades.Thefirst international conference on diver distraction and inattention was held in2009[11].A number of auto companies have already installed simple function driver fatigue monitoring systems in their high-end vehicles.However, there is still a great need to develop a more reliable and fully functional DIMS using cost-efficient methods in a real driving context.It is believed that the development of signal processing and computer vision techniques will attract more attention to the study of thisfield in the coming years.With the intention of benefiting individuals or groups interested in or are about to enter thisfield,this paper gives a comprehensive review of the state of the knowledge on driver inattention.It thus provides a clear view of the previous achievements and the issues that still need to be considered.
This paper is organized as follows.We introduced the driver inattention concept in Section I.Next,the effects of driver distraction and fatigue on driving performance are prented in Section II.Becau some commercial products relative to inattention detection have emerged on the market in recent years,Section III is devoted to reviewing the products. Section IV prents a detailed review of the scientific rearches on inattention detection.The followingfive types of measures for inattention detection are prented in this ction:
1)subjective report measures;
2)driver biological measures;
3)driver physical measures;scrutinize
4)driving performance measures;
5)hybrid measures.
After a discussion in Section V,we prent a conclusion and propo some areas for future study in Section VI.
II.D ISTRACTION AND F ATIGUE E FFECTS ON D RIVING
仲由B EHAVIORAL P ERFORMANCE
This ction concentrates on how distraction and fatigue affect a driver’s behavior and driving performance.Explor-ing the effects could provide uful information for the development of real-time distraction and fatigue detection algorithms.
A.Effects of Distraction
Performing a cognitively demanding task while driving influ-ences both the driver’s visual behavior and driving performance (as indicated by braking behavior).
1)Driver Behavior Patterns:With an increa in the cog-nitive demand,many drivers changed their inspection patterns on the forward view.Angell et al.[12]indicated that the eye-glance pattern could be ud to discriminate driving while performing a condary task from driving alone and could be ud to discriminate high-from low-workload condary tasks. More facts associated with cognitive distraction driving can be found in[13]and[14]:Drivers narrowed their inspection of the outward view and spent more time looking directly ahead.They reduced their inspection of the instruments and mirrors and reduced their glances at traffic signals and the area around an interction.Rantanen and Goldberg[14]found that the visual field shrank by7.8%during a moderate-workload counting task
and by13.6%during a cognitively demanding counting task. Drivers had fewer saccades per unit time,which was consistent with a reduction in glance frequency and less exploration of the driving environment,and in some cas,drivers completely shed the tasks and did not inspect the areas at all[15]. Hayhoe[16]showed links between eye movement(fixation, saccade,and smooth pursuit),cognitive workload,and dis-traction.Fixations occur when an obrver’s eyes are nearly stationary.Saccades are very fast movements that occur when visual attention shifts from one location to another.Smooth pursuits occur when an obrver tracks a moving object such as a passing vehicle.Saccade distance decreas as task complex-ity increas,which indicates that saccades may be a valuable index of mental workload[17].In contrast,the amount of head movement incread when cognitive loads were impod.It is believed that this condition is a compensatory action by which a driver attempts to obtain a widerfield of view[18]. Miyaji et al.[18]propod that the standard deviations of eye movement and head movement could be suitable for detecting the states of cognitive distraction in subjects.Both cognitive and visual distractions caud gaze concentration and slow saccades when drivers looked at the roadway,and cognitive distraction incread blink frequency[19].Liang and Lee[19] found that visual distraction resulted in frequent long off-road glances.A report from the Safety Vehicle Using Adaptive In-terface Technology(SA VE-IT)program showed that eyes-off-road glance duration,head-off-road glance time,and standard deviation of lane position(SDLP)are good measures of visual distraction[20].
2)Other Physiological Respons:When cognitive loads (conversation or arithmetic)were impod on subjects,pupil dilation occurred by the acceleration of the sympathetic nerve [18].The average heart rate also incread by approximately 8beats per minute.However,the average value of the heart rate [R-to-R interval(RRI)]decread under the same situation[18]. Itoh[21]pointed out that performing a cognitively distracting condary ,talking or thinking about something)dur-ing driving caud a decrea in the driver’s temperature at the tip of the no,and this effect was reproducible.It was reported in[22]that a considerable and consistent skin temperature increa in the supraorbital region could be obrved during cognitive and visual distractions.Berka et al.[23]found that the electroencephalography(EEG)signal also contained infor-mation about the task engagement level and mental workload.
3)Driving Performance:Significant changes were ob-rved in a driver’s vehicle control as a conquence of per-forming additional cognitive tasks while driving.Ranney[24] found that distraction may be associated with laps in vehicle control,resulting in unintended speed changes or allowing the vehicle to drift outside the lane boundaries.Zhou et al.[25] found the influences on the lane-changing behavior when a condary task was performed,which included a reduction in the frequency of the checking behavior(check a side mirror or speedometer),a delay in the checking beh
avior,and a longer time to perform the checking behavior.Carsten and Brookhuis[26]found that the effects of cognitive distraction on
driving performance considerably differed from the effects of visual distraction.Visual distraction affects a driver’s steering ability and lateral vehicle control,whereas cognitive distraction affects longitudinal vehicle control,particularly car following. Liang and Lee[19]also found that cognitive distraction made steering less smooth but improved lane maintenance.In addi-tion,Liang and Lee[19]found that steering neglect and over-compensation are associated with visual distraction,whereas undercompensation is associated with cognitive distraction. Overall,visual distraction interferes with driving performance more than cognitive distraction.One apparently anomalous finding is that,when condary task cognitive demands in-cread,a driver’s lateral control ability was found to improve [26].Harbluk et al.[13],[15]found an incread incidence of hard braking associated with cognitive distraction driving. B.Effects of Fatigue
When a driver is fatigued,certain physical and physiological phenomena can be obrved,including changes in brain waves or EEG,eye activity,facial expressions,head nodding,body sagging posture,heart rate,pul,skin electric potential,grip-ping force on the steering wheel,and other changes in body activities.
1)Driver Behavior Patterns:Eskandarian et al.[27]found that the following actions were correlated with fatigue.
1)Drivers exhibited a reflexive head nod after checking the
side mirrors.
2)The head motions were significantly less frequent.
3)The number of times that the drivers touched or scratched
their chin,face,head,ears,eyes,and legs significantly incread.
4)Drivers were inclined to turn their head to the left to
relieve muscular tension in the neck.
5)Eye-blinking activity radically incread.
6)Episodes of yawning were more frequent.
7)Drivers tended to adopt more relaxed hand positions on
the steering wheel.
In particular,for eye-blinking patterns,PERCLOS[28], which is the percentage of time that the eye is more than 80%clod,is one of the most widely accepted measures in the scientific literature for drowsiness detection.It has been validated using both EEG data and subjective evaluation.
2)Other Physiological Respons:The activity of a low-frequency EEG ranging from0to20Hz has a significant re-lationship with sleepiness.The spectral analysis of an EEG that shows the transition from wakefulness to sleep can be described as a shift toward slower EEG frequencies.In the alert condition, the appearance ofβactivity is common in the EEG.αactivity is also normally found in the occipital regions(O1and O2)in the awake and relaxed conditions.When a driver gets drowsy, a burst ofαactivity can often be en in the central regions of the brain(C3and C4).However,some people do not show any αactivity.As the driver gets drowsier,theαactivity is replaced byθactivity.Whenδactivity occurs in the EEG,the driver is no longer awake,which is an indicator of deep sleep[29].
学习考察报告3)Driving Performance:It has been reported that sleep-deprived drivers have a lower frequency of steering reversals (every time the steering angle cross zero degrees)[30], a deterioration of steering performance[31],a decrea in the steering-wheel reversing rate[32],more frequent steer-in
g maneuvers during wakeful periods,no steering correction for a prolonged period of time followed by a jerky motion during drowsy periods[33],low-velocity steering[34],large-amplitude steering-wheel movements,and large standard devi-ations in the steering-wheel angle[35].Zhong et al.[36]found that when drivers had a fatigued status,the steering-wheel angle and vehicle tracking became irregular,and the range of deviation greatly incread.Several rearchers found that the lane-tracking ability decread as the time on the task incread [31].Variables such as the times of lane departures,SDLP,and maximum lane deviation were found to highly be correlated with eye closures[37].The mean square of lane deviation,mean square of high-pass lateral position,and SDLP showed good potential as drowsiness indicators[38].
Dingus et al.[34]found that the yaw deviation variance and the mean yaw deviation(calculated over a3-min period) showed some promi as drowsiness indicators.However,no strong correlations between drowsiness and braking or acceler-ation were found in[34]and[39].Generally,vehicle speed vari-ability has not shown any strong correlation with drowsiness [39].However,some reports found that the standard deviation of speed incread from the third driving hour,with a time interval of45min[40].
III.C OMMERCIAL P RODUCTS AND A CTIVITIES FOR
D RIVER I NATTENTION D ETECTION
A.Auto Companies
Several famous auto companies are currently conducting rearches on driver inattention monitoring systems,including Toyota,Nissan,V olvo,Mercedes-Benz,and Saab.
Saab’s Driver Attention Warning System[41],[42]is a project designed to counter the following two most common caus of road accidents:1)driver drowsiness and2)distrac-tion.The system utilizes two miniature infrared(IR)cameras: one camera installed at the ba of the driver’s A-pillar and the other camera at the center of the main fascia,which are focud on the driver’s eyes.It also utilizes the SmartEye[43] software to get accurate eyelid,gaze,and head orientation infor-mation.In their algorithm,the driver’s eye blinking frequency is measured.If a pattern of long-duration eyelid closures is detected,it indicates the potential ont of drowsiness.A three-level warning interface was designed for drowsiness detection. This condition starts with a chime sound and text message,then it moves on to a spoken message,andfinally,a stronger warning tone audio message is persistently delivered until the driver press the ret button.As soon as the driver’s gaze moves away from what is defined as the“primary attention zone”—the central part of the windshield in front of the driver—
a timer starts counting.If within2s of the timer being triggered the driver’s eyes and head do not return to the“straight ahead”position,it is considered a distraction.In a ca that involves peripheral tasks such as looking in the rear-view mirror,a side
mirror,or turning a corner,the timer’s elap time becomes longer.Once the driver distraction has been detected,a at vibration signal will be issued to warn the driver.However, there is no report about the robustness of this system dur-ing daytime and nighttime driving under different kinds of weather conditions,providing no driver status ground truth as a reference.
Toyota developed their Driver Monitoring System in2006 for the latest Lexus models.This system features a camera, which us near-IR technology,mounted on top of the steering column cover.It monitors the exact position and angle of the driver’s head while the vehicle is in motion.If the Advanced Precrash Safety system detects an obstacle ahead,and at the same time,the Driver Monitoring System establishes that the driver’s head has been turned away from the road for very long,the system automatically activates precrash warnings. If the situation persists,the system can briefly apply the brakes to alert the driver[44].In2008,the Toyota Crown System went further.It can detect if drivers become sleepy by monitoring their eyelids.Toyota’s solution combines driver face orientation and environmental obstacle detection to deter-mine accident potential and utilizes eyelid
activity to identify drowsiness.
In the spring of2009,Mercedes-Benz introduced Attention Assist into its ries production[45].Attention Assist works byfirst obrving a driver’s behavior and then us this infor-mation to create a unique driver profile.During operation,a ries of tests continually monitor the driver input in relation to this profile,and in the event that a deviation is encountered, the system then determines whether the deviation is a result of fatigue.If it is,Attention Assist both visually and audibly alerts the driver that it is time to take a break.The factors taken into account to determine a driver’s profile include the speed,lon-gitudinal and lateral acceleration,angle of the steering wheel, the way that the indicators and pedals are ud,certain driver control actions,and even various external influences such as a side wind or an uneven road surface.The Attention Assist sys-tem only us vehicle parameters to determine driver drowsi-ness,which requires no additional hardware tup.However, this system needs to establish individual profiles for different drivers,which would affect the acceptance of the system in real life.
In2007,V olvo Cars introduced Driver Alert Control to alert tired and nonconcentrating drivers[46].With the idea that the technology for monitoring a driver’s eyes is not yet sufficiently mature and human behavior varies from one person to another, V olvo Cars developed the system b
ad on the car’s progress on the road.It is reported that Driver Alert Control monitors the car’s movements and asss whether the vehicle is driven in a controlled or uncontrolled way.It can also cover situations where the driver focus too much on his/her cell phone or children in the car,thereby not having full control of the vehicle.
B.Other Commercial Products and Activities Technological approaches have continued to emerge in recent years and hold promi for detecting and monitoring dangerous levels of driver inattention.Although many of the projects are now in the development,validation testing,or early imple-mentation stages,some companies can provide usable devices or prototypes to give information about driver behavior.For nonintrusive measurement,the devices mainly utilize video cameras and computer vision technologies.
Attention Technology,Inc.has designed and developed the DD850Driver Fatigue Monitor(DFM):the only real-time on-board drowsiness monitor that is currently tested in an extensive field operational test.The DFM is a video-bad drowsiness detection system that works by measuring slow eyelid closure. It is designed to mount on a vehicle’s dashboard just to the right of the steering wheel and provides a continuous real-time mea-surement of eye position and eyelid closure[47].In particular, DFM estimates PERCLOS to determine drowsiness,which is the proportion of time that the eyes are
clod80%or more over a specified time interval.DFM us a structured illumination approach to identify the driver’s eyes.This approach obtains two concutive images of the driver using a single camera. Thefirst image is acquired using an IR illumination source that produces a bright-pupil image.The cond image us an IR illumination source at a different wavelength to produce an image with dark pupils.The two images are esntially identical,except for the brightness of the pupils in the images. The third image calculates the difference between the two images,enhancing the bright eyes and eliminating all image features,except for the bright pupils.The driver’s eyes are identified in this third image by applying a threshold to the pixel brightness.The bright-pupil effect utilized by DFM is a simple and effective eye-tracking approach for pupil detection bad on a differential lighting scheme.However,the success of the bright-pupil technique strongly depends on the brightness and size of the pupils,which are often functions of face orientation, external illumination interference,the distance of the subject from the camera,and race.For real-world in-vehicle applica-tions,sunlight can interfere with IR illumination,reflections from eyeglass can create confounding bright spots near the eyes,and sunglass tend to disturb the IR light and make the bright-pupil phenomenon appear very weak.
Delphi believes that computer vision offers the most direct method for detecting the early ont of sle
epiness and distrac-tion,and it is also en as an excellent platform to be shared with other vision-bad driver assistance applications in the future.They integrated two ,the ForeWarn Drowsy Driver Alert and the ForeWarn Driver Distraction Alert,into a comprehensive Driver State Monitor(DSM)[47].DSM is a computer vision system that us a single camera mounted on the dashboard directly in front of the driver and two IR illumination sources.Upon detecting and tracking the driver’s facial features,the system analyzes eye closures and head po over time to infer the fatigue or distraction level.It provides an extended eye-closure warning for closures longer than2.5s and provides an extended distraction warning for nonforward gaze states in excess of2.5s.The fatigue detection algorithm predicts A VECLOS,the percentage of time that the eyes are estimated to be fully clod over a1-min interval.Becau this approach is a less-complex measure of drowsiness than PERCLOS,it permits the u of an automotive-grade data
processor,in contrast to the high-grade PC processor required for PERCLOS.
Seeing Machines is engaged in the rearch,development, and production of advanced computer vision systems for rearch on human performance measurement,advanced driver-assistance systems,and transportation[48].Their signa-ture ,faceLAB,provides head and face tracking, as well as eye,eyelid,and gaze tracking for human subjects, using a noncontact video-ba
d nsor.faceLAB provides com-prehensive blink analysis and PERCLOS asssment,including the delivery of raw data on the details of eyelid behavior. Instead of using traditional corneal reflection techniques,input is obtained using a stereo camera pair.Seeing Machines face-LAB has extensively been employed as a PC-bad rearch tool.Although the device reportedly works very well in a simulator environment,the numerous challenges faced in a real driving environment prevent it from robustly working.Seeing Machines also provides another product:Driver State Sensor (DSS).It consists of one camera,two IR light-emitting diode (LED)illuminators,and one special computing and commu-nication unit.The goal of DSS is to detect driver fatigue by analyzing eyelid activity.
Smart Eye AB is another company that provides computer-vision-bad software that detects human face/head movement, eye movement,and gaze direction[43].Their , Smart Eye Pro3.0,is a machine vision system that estimates head po using a simple and robust method bad on tracking individual facial features and a3-D head model.Although the face is tracked,the gaze direction and eyelid positions are determined by combining image edge information with3-D models of the eye and eyelids.One major advantage is that eye and head tracking can continue,although one camera is fully occluded or otherwi nonoperational.This approach also allows for large head motions(translation and rotation).Smart Eye has not developed an algorithm that monitors drowsines
s. SensoMotoric Instruments GmbH(SMI)[49]is a German company who ,InSight,can measure head posi-tion and orientation,gaze direction,eyelid opening,and pupil position and diameter.InSight us a sampling rate of120Hz for head po and gaze measurement,120Hz for eyelid closure and blink measurement,and60Hz for combined gaze,head po,and eyelid measurement.It also provides PERCLOS information for drowsiness detection.It is a computer-bad system and needs ur calibration.
IV.C URRENT M ETHODS OF D ETECTING
D RIVER I NATTENTION
In the scientific literature,the followingfive main types of measures for inattention detection are commonly ud:
1)subjective report measures;
2)driver biological measures;
3)driver physical measures;决定的英文
4)driving performance measures;
教师培训班5)hybrid measures.
With the exception of subjective report measures,the measures are bad on nonlinear modeling techniques.In this ction,we will briefly review the most common nonlinear modeling techniques.Then,the rearches on thefive main types of measures will be explored.Finally,the extraction of physical signals from a driver by image processing will be discusd at the end of this ction,becau driver physical measures offer distraction detection through eye gaze monitor-ing and fatigue detection through eye gaze,blink,head,and mouth tracking.
A.Nonlinear Modeling Techniques
Human cognition can hardly be reprented by a linear model.Hence,nonlinear modeling techniques are greatly adopted in the driver inattention detection area.Nonlinear mod-eling with machine learning techniques can extract information from noisy data and do not require prior knowledge before training.Some mechanisms in machine learning can avoid overfitting for nonlinear modeling,producing more robust and general models than traditional learning ,logistic regression),which only minimize training error.
Artificial neural networks(ANNs)have been studied and utilized in numerous scientific and engineerin
gfields.One of the main advantages of ANNs is that they infer solutions from data with no prior knowledge of the patterns in the data, i.e.,they empirically extract the patterns even if the equation between the inputs and outputs does not exist.This charac-teristic is very important,becau in most practical cas,the exact input–output relationship is difficult to establish.ANNs also have the ability to ,they respond with a reasonable accuracy to patterns that are broadly similar to the original training patterns),which is very uful,becau real-world data are noisy,distorted,and often incomplete.ANNs are nonlinear,which allows them to more accurately solve some complex problems than linear techniques[27].
The fuzzy inference system(FIS)is famous for its well-known linguistic concept modeling ability.The fuzzy rule ex-pression is clo to an expert natural language.A fuzzy system then manages the uncertain knowledge and infers high-level behaviors from the obrved data.On the other hand,becau it is a universal approximator,FIS can be ud for knowledge induction process[50].
The support vector machine(SVM)is bad on the statistical learning technique and can be ud for pattern classification and the inference of nonlinear relationships between variables.This method has successfully been applied to the detection,verifica-tion,and recognition of faces,objects,handwritten characters and digits,text,speech,and speakers,along with the retrieval of information and images[51]
.The learning technique of the SVM method makes it suitable to measure the cognitive states of humans.SVMs can generate both linear and nonlinear models and can compute the nonlinear models as efficiently as the linear ones.Given a t of input data,this method first transforms the input domain through a kernel and then looks for a hyperplane in the transformed domain that parates the data with minimum error and maximum gain.Finally,the hyperplane is transformed back to the input domain to obtain the decision boundaries,which may potentially be nonlinear. AdaBoost is a learning algorithm that us the pattern-recognition algorithm called boosting[52].Its advantages
