TECHNICAL PAPER SUBMISSION

Paper title: ADVANCED DRIVER MONITORING- the AWAKE project -

PRESENTING AUTHOR Family Name: Bekiaris
Title (Prof/Dr/Mr/Mrs/Ms/Miss): Dr. Given Name: Evangelos
Department: Hellenic Institute of Transport Organisation: Centre for Research and Technology Hellas
Address: 6th km Charilaou Thermi Road
City: Thermi, Thessaloniki State/Province: Postcode: 57001 Country: Greece
Telephone: 30/31/481265 Facsimile: 30/31/481269
(country/area codes) (country/area codes)
Email: [email protected]

Co Authors: Wevers, Amditis

Title (Prof/Dr/Mr/Mrs/Ms/Miss): Mr. Given Name: Kees
Department: Organisation: Navigation Technologies B.V.
Address: De Waal 15
City: Best State/Province: Postcode: NL-5684 PH Country: The Netherlands

Email: [email protected]
Telephone: +31.499.331.585 Facsimile: +31.499.331.410
(country/area codes) (country/area codes)
Title (Prof/Dr/Mr/Mrs/Ms/Miss): Dr. Given Name: Angelos
Department: Organisation: Institute of Communication and Computer Systems
Address: 9, Iroon Politechniou str.
City: Zografou State/Province: Postcode: 15773 Country: Greece
Telephone: +30/17722398 Facsimile: +30/1 7723557
(country/area codes) (country/area codes)
Email: [email protected]
Title (Prof/Dr/Mr/Mrs/Ms/Miss): Given Name:
Department: Organisation:
Address:
City: State/Province: Postcode: Country:
Telephone: Facsimile:
(country/area codes) (country/area codes)
Email:
Title (Prof/Dr/Mr/Mrs/Ms/Miss): Given Name:
Department: Organisation:
Address:
City: State/Province: Postcode: Country:
Telephone: Facsimile:
(country/area codes) (country/area codes)
Email:
Title (Prof/Dr/Mr/Mrs/Ms/Miss): Given Name:
Department: Organisation:
Address:
City: State/Province: Postcode: Country:
Telephone: Facsimile:
(country/area codes) (country/area codes)
Email:
 ADVANCED DRIVER MONITORING THE AWAKE
PROJECT
Dr. Evangelos Bekiaris
CERTH/HIT, 6th km. Charilaou Thermi Road, 57001 Thermi, Greece
Tel. +30 31 498265, Fax +30 31 498269, e-mail: [email protected]

Kees Wevers
Navigation Technologies B.V., De Waal 15, NL-5684 PH, The Netherlands
Tel.: +31.499.331.585, Fax: +31.499.331.410
E-mail: [email protected]

Dr. Angelos Amditis

Institute of Communication and Computer Systems of the National Technical University of
Athens, Microwave and Optics Laboratory
9, Iroon Politechniou Str., 15773 Zografou, Greece
Tel. ++301 7722398, Fax: ++301 7723557
E-mail: [email protected]

ABSTRACT

Recent research indicates that driver hypovigilance (or under-awakeness) is a major cause of road
accidents. The objective of the AWAKE project, co-funded by the IST programme of the European
Commission, is the development of an unobtrusive and reliable in-vehicle system to monitor the
driver and the environment, for real-time detection of hypovigilance, based on multiple parameters.
Continuous (instead of discrete) event related driver monitoring, effective system personalisation
based on driver characteristics, and consideration of the actual traffic situation will enhance the
reliability of the system and minimise the false alarm rate. In case of hypovigilance the system will
provide an adequate warning to the driver. Several warning levels will be used, depending on the
(estimated) level of driver hypovigilance, and the estimated level of traffic risk. This paper presents
the general concept, the design process and the modules of the AWAKE system.

INTRODUCTION

A major focus of research over the last few years has been driver hypovigilance as a cause of
road accidents. National Transportation and Safety Board of US has during the 1990s paid attention
to driver fatigue as one of the most important causes of road accidents [18]. 10-20% of all accidents
is related to driver fatigue [19]. More precisely:

[2] found that fatigue and/or drowsiness of the driver caused around 30% of accidents in French
highways in the period 1979-1994, whereas about 40% of fatal accidents on US highways are
sleep-related [3].
1% to 10% of all accidents in the U.S.A. seem to be directly related to sleepiness [4].
Regarding heavy vehicles crashes [5] estimated that in the USA fatigue-related crashes
constitute 0.71%-2.7% of all crashes involving trucks and 15% to 36% of all crashes fatal to the
truck driver. [3] estimates that fatigue is a factor affecting 30-40% of heavy truck crashes in US.
[6] and [7] establish that a good detection of fatigue alone could concern between 40% and
2
 60% of the crashes with one vehicle and 37% of truck drivers fatalities.
Expected involvement in such accidents of trucks is 4.5 times greater than for passenger
vehicles due to exposure, operational life and night driving [7].

Furthermore, accidents related to driver hypovigilance are more serious than other types of
accidents. An impaired driver will not take evasive action prior to a collision, and if a cruise control
is used, the vehicle will keep its speed until a major impact. [9] concludes that the reductions in
traffic crash losses from reducing crashes attributable to driver impairment far exceed reductions
from any other potential countermeasure.

The circumstances in which drowsiness-related accidents usually happen should be taken into
account, namely late-night hours, with a smaller peak in the mid-afternoon. Young drivers have no
increased risk during the afternoon. Drivers over 45 on the other hand have fewer night time
crashes, with a peak at 7 a.m., and are more likely to have such crashes during the mid-afternoon
[11]. For this reason, the age groups specifically prone for such accidents are being focused within
AWAKE. For example, [11] estimates that drivers younger than 30 account for almost two thirds of
drowsy driving crashes, despite counting for only about one-fourth of licensed drivers. 20 is the age
of peak occurrence of drowsy-driving crashes, whereas both the 18-24 and 25-39 age groups are
over-represented in fall-sleep crashes. Data like the above will guide the user groups synthesis of
the project pilots.

In addition, the problem might be also directly related to the introduction of various Advanced
Driver Assistance Systems (ADAS) in the next years in the market. Automated or even assistive
driving systems may also induce fatigue and stress to the driver caused by prolonged driving under
monotonous driving conditions. In other words, cases of driver hypovigilance may be well
enhanced due to the introduction of ADAS. And yet, ADAS technology may well be the right
answer to fight this same problem .

Thus, the objective of AWAKE is to increase traffic safety by reducing the number and the
consequences of traffic accidents caused by driver hypovigilance. In order to achieve this
objective, AWAKE intends to develop an unobtrusive, reliable system, which will monitor the
driver and the environment and will detect in real time hypovigilance, based on multiple
parameters. The system will achieve enhanced reliability and minimised false alarm rate, by
supporting continuous, instead if discrete, event-related driver monitoring, strong system
personalisation to driver characteristics and traffic situation awareness. In case of
hypovigilance, the system will provide an adequate warning to the driver, with various levels of
warnings, according to the estimated drivers hypovigilance state and also to the estimated level of
traffic risk. This system will operate reliably and effectively in all highway scenarios .

STATE OF THE ART

Various prototypes and systems for monitoring driver impairment have been developed until now,
without any outstanding market success. Most of them either base their detection on single driver
characteristics (i.e. eyelid movements, eye closure, steering grip force), on physiological
measurements of the driver (such as EEG, EOG, ECG, muscle activity) or on behavioural
characteristics (such as vehicle speed, lane position, etc.). Very few attempts have been made on
combining all above indexes in a complementary way. Yet, these few attempts have shown
promising results, even if they have not as yet solved the problem in its entity.

The most promising efforts so far include:

3
 Nissan anti-drowsiness system that monitors drivers eyelid movements and in case drowsiness
is detected warns the driver by audio signal and at a second stage releases menthol scent and
cold air to awake the driver. It also includes an automatic braking option.
Mitsubishi Drivers View Detector uses two dash-mounted cameras to monitor and analyse eye
position and blinking. If drowsiness is detected, a warning sound is generated.
Within SAVE EU project (System for effective Assessment of driver state and Vehicle control
in Emergency situations - TR 1047) an eyelid sensor has been developed for driver
hypovigilance detection. This sensor has high detection rate under the condition that the driver
does not wear glasses. Furthermore, detection rate is influenced by environmental lighting
conditions and even by the drivers driving style (i.e. when putting the hands on the steering
wheel). Still, under optimal conditions, detection rate of up to 95% can be achieved [10].
AWAKE is the follow-up of SAVE project.
[11] has evaluated various techniques for ocular measurements as the basis for alertness
management. They have compared two EEG algorithms, a head tracker device, two wearable
eye-blink monitors with a video-based scoring of eye closure by trained observers. The results
suggest that the MICRODAS system using the PERCLOS algorithm is the best of these
technologies in simulation studies and can be used to detect hypo-vigilance under the condition
that it will be automated in a computer algorithm. The false alarm rate of the algorithm has been
reported to be 0.5% and the miss-rate 19.73% by [12]. A follow-up [13], however, using real
world driving data was inconclusive in relating the algorithm results to the prediction of driver
hypovigilance. They have found a false alarm rate of 1.6% but a miss-rate of 37.2%.
The Spanish ADS (Anti-Drowsiness System) detects driver drowsiness by monitoring steering
grip pressure at the steering wheel, warns the surrounding traffic (by blinking the headlights and
acoustic alarm) and automatically cuts off the vehicles fuelling to stop it, in case driver
drowsiness is detected. No reliability or market data on this system are available.
Dozers alarm from Australia detects drivers drowsiness by measuring the inclination angle of
drivers head. No reliability or market data are available for this system too.
A steering grip detector developed within SAVE project (TR 1047) had a rather low detection
rate (60%), when used as stand-alone sensor for drowsiness detection. Furthermore, a head-rest
sensor developed in the same project seemed to be useful to detect sudden loss of control by the
driver (i.e. fainting, heart attack) but has low correlation to vigilance detection [10].
Head pose and gaze-direction tracking system developed by an Australian University (ANU)
and Volvo [23], estimates fatigue through measuring head-pose, gaze-direction, eye closure and
blinking. The system is not yet in the market (is announced for 2001 but without specified
price). Its setup process requires 10 minutes for an environment with controlled lighting, but
in-vehicle lighting conditions are rapidly and constantly changing. It still requires a few markers
on the drivers face.
- SAMOVAR Project (V2007 - Safety Assessment Monitoring On-Vehicle with Automatic
Recording) concentrated on low cost in-vehicle electronic system for recording vehicle data
and driver behaviour, in relation to drivers performance of the driving task (i.e. speed choice,
traffic signal violation, lane keeping), but did not result in any specific correlation of them to
drivers hypovigilance.
- DETER (V2009 - Detection, Enforcement and Tutoring for Error Reduction) project has been
based on DRIVE I projects AUTOPOLIS (V1033 - Automatic Policing Information Systems),
DREAM (Feasibility of Driver Monitoring) and GIDS (V1041 - Generic Intelligent Driver
Support Systems). All projects targeted driver vigilance detection by monitoring drivers
behaviour, concerning speed, headway, violation of traffic law. For this they used strategies
like electro-encephalograms, electro-oculograms, off-line video camera analysis of driver's
face, etc. No actual system resulted from them.
- Daimler Chrysler [14] has developed a detection algorithm that uses lateral position, steering
wheel angle and longitudinal speed data, jointly analysed to detect drivers drowsiness. This

4
 algorithm has been validated in simulator and field studies with drivers of private cars and
heavy vehicles. This system has been proven feasible but is not fully developed yet.
- SafeTRAC [24] is a new system from USA (just marketed), which claims to effectively monitor
driver drowsiness in over 97% of all highway driving conditions with less than 1 false alarm
every 8 hours of driving. It is based on a lane tracking system. The system requires white lane
markers to operate and does not operate under night time rain conditions. Its reported false
alarm level of 1 false alarm every trip (a professional drives typically 8 hours per journey) is
still too high and the environmental restrictions seem too severe.

An abundance of systems have tried to detect on-line driver monitoring by the measurement of
drivers physiological measurements, such as EEG, ECG, EOG signals, cerebral, cardiac and
pulmonary activities. As two typical examples the following are mentioned:
Toyota has developed an anti-drowsiness system, that detects driver drowsiness by
physiological driver measurements through a wrist device at drivers left hand (like a watch)
and warns the driver by alarming sounds (first stage), vibration of the drivers seat (second
stage) and even automatic braking (third stage). Drivers wiring has been reported to be
particularly disliked by the users.
Atlas Researches from Israel has developed a family of relevant monitors, called NOVAlert,
that employ electromyography for monitoring drivers wrist activity (by a wireless wrist
detector) and warning the driver by vibratory stimulus, to increase muscle activity if he/she
slows the pace or reduces isometric levels below a relevant criterion.

Such systems however target more driver alertness maintenance than monitoring.

AWAKE OBJECTIVES AND INNOVATIONS

From the above cases as well as from analysis of the most significant drawbacks mentioned in the
literature, AWAKE Consortium has concluded that:

A multi-sensor approach, combining drivers physiological and behavioural (traffic task related)
parameters is the only feasible way of successfully filtering driver, traffic, environment-caused
disturbances and noise to any single sensor.
Driver wiring and the use of obtrusive sensors has to be excluded due to their low user
acceptance. Also, systems requesting often feedback by the driver should be avoided for the
same reason (i.e. to perform a task in order to prove he/she is awake or to keep the driver awake
for longer).
There is a necessity to develop a continuous diagnostic algorithm instead of a discrete one,
allowing for incremental learning and sufficient personalisation to the particular driver
characteristics.
There is a need for on-line environmental condition adaptation of the detection algorithm, based
on the actual traffic conditions.
Driver warning should be modular and multimodal, to force the driver to understand the danger
and react to it, without causing him/her panic. If the surrounding traffic situation permits it, high
levels of warning should be avoided to reduce users disturbance.
System intervention (for autonomous vehicle control) is feasible but still too far from the market
(due to the associated high failure risk of any subsystem and its enormous cost), and thus can not
be effectively supported yet.
A uniquely defined unit will not be able to cover all types of users and problems. For example,
Sleep Apnea Syndrome (SAS) patients may accept more strict detection and warning strategies

5
 than professional drivers, who in turn would tolerate much more than passenger vehicles drivers,
and especially young ones. Thus, a modular and configurable system should be targeted.

AWAKE will strive to meet the above requirements for an optimum detection system by a series of
innovations in its specifications, design and development, prominent among them being:

1. Development of a multi-sensor system, including the most promising detection sensors

developed so far, such as an eyelid camera, a steering grip sensor, a lane tracker and other
vehicle-related parameters monitoring sensors. For this the best existing sensors in Europe
(all available to the Consortium) will be further optimised.
2. Follow-up in parallel of the most promising stochastic and deterministic (knowledge-based)
approaches for data fusion of the above sensor data.
3. Extension of the current diagnostic methods to take into account continuous diagnostic
capabilities (incremental learning).
4. On-line personalisation of the diagnostic algorithm.
5. Introduction of ambient intelligence to the diagnostic module, according to the traffic
environment and drivers attention to it (through gaze analysis).
6. Development of a warning strategy that combines acoustic, visual and haptic elements and is
parametric to drivers vigilance state as well as to the estimated traffic situation.
7. Emphasis on cost-effective and viable solution to the market, using to the maximum extent
readily available sensors and other ADAS subsystems.
8. Emphasis on legal, standardisation and insurance issues, to guarantee unobstructed
marketing of the developed system.
9. Parametric specifications that match the particular preferences of the target user cohorts
(such as young drivers, professional / heavy vehicle drivers, shift workers and people
suffering from sleep disorders) and the whole spectrum of application fields (small to luxury
passenger cars, heavy vehicles).

THE AWAKE SYSTEM

The system will consist of the following modular components:

A Hypovigilance Diagnosis Module (HDM ) that will detect and diagnose driver hypovigilance in
real-time. Based on an artificial intelligence algorithm this module will fuse data from on-board
driver monitoring sensors (eyelid and steering grip data) and data regarding the driver's behaviour
(lane tracking, gas/brake and steering position data). The HDM will be adapted to the specific
driving characteristics of the user by continuous driver monitoring and expert-based adaptation. The
goal is to achieve a (correct) diagnosis level over 90% and a false alarm rate below 1% in all
highway scenarios.

A Traffic Risk Estimation Module (TREM ) that will assess the traffic situation and the involved
risks. It will match, following a deterministic approach, data from an enhanced digital navigational
map, a positioning system, an anticollision radar, the odometer, and a driver's gaze direction sensor.
This module is not designed as a complete new system to estimate traffic risk, but rather as an
expert combination of existing ADAS technology. The output of this module will be used by the
HDM to re-assess the state of the driver, and by the Driver Warning System to determine the
adequate level of warning.

A Driver Warning S ystem (DWS) that will use acoustic, visual and haptic means. The module
will use inputs from the HDM and the TREM to determine the adequate warning level for a certain
6
situation.

A Hierarchical Manager (HM ) which will be able to perform self-diagnosis, and which will
co-ordinate the other system components.

M entioned system components and sensors will be integrated in one single unit (the AWAKE unit)
suitable for use in real-life automotive applications (in terms of cost, size, weight and robustness).
In the project three prototype systems will be developed, and tests will be performed with a middle
and an upper class passenger car, and with a heavy vehicle demonstrator, so as to cover all relevant
applications areas.

EXPECTED IMPACTS OF AWAKE

Traffic crashes constitute one of the largest public health problems in industrialised countries. In the
USA almost half of the deaths of 19-years-olds are caused by traffic crashes, and the total number
of pre-retirement years of life cost because of traffic crashes is approximately equal to deaths
caused by the combined effects of the two leading diseases, cancer and heart diseases [9].

Over 30% of accidents may have as primary or secondary cause the driver impairment, due to a
variety of reasons. Also. almost 30% of accidents could be avoided by means of reducing the driver
related reaction time by just 0.5 sec (through warning by AWAKE system).

The above reasoning allows to believe that a successful application of the AWAKE concept will
have important impact on road safety, namely it may seriously reduce accident rate. Still, this is far
from certain. Although no model exists which can predict the actual safety effect of a new
intervention, the results of previous interventions suggest some fairly stable patterns. Thus, better
brakes and handling, poor-weather vision and drowsiness detection are expected to lead to faster
driving, faster cornering, faster speeds under low visibility, and longer-duration driving [9].

Hence, only after the behavioural effects of AWAKE system have been studied (in its pilots) the
actual impact to traffic safety will be estimated. If we do not consider possible negative effects (i.e.
from over-reliance to the system), hoping to avert them by proper system implementation, we may
result in erroneous assumptions.

On the other hand, the introduction of other ADAS into traffic (such as obstacle avoidance, vision
enhancement, route guidance and lane keeping) may well induce subjective fatigue and stress or
boredom to the drivers, leading to even more monotonous driving conditions. Thus, the need for
driver monitoring and drowsiness detection systems will be further enlarged.

The implementation of AWAKE system would have other benefits too. In the [Intelligent Speed
Adaptation pamphlet of AVV, 1997] it is mentioned that an accident reduction by 21% means a
reduction in fuel consumption and carbon dioxide emissions by 11% while nitrogen oxide
emissions would be reduced even more, by 15%, as accidents have as indirect impact traffic jams
and a temporal decrease of the networks capacity. Hence, an eventual accident reduction by
AWAKE system will also be accompanied by positive environmental impact.

The total cost of crashes in EU is 50 billions Euros [High Level Group Report, 1991]. Worldwide,
more than half a million people are killed each year in traffic crashes. Approximately one person in
200 in the worlds population dies from injuries received in traffic crashes while about 15 million
people per year are injured in traffic crashes worldwide. This means that the average citizen of the
7
world has about one in seven chance of being injured in a traffic crash sometime during ones life.
Furthermore, it has been estimated [NHTSA, 1989] that the average vehicle has about a 20%
probability of being involved in some type of crash per year.

More specifically, according to [4] sleep-related accidents in the USA are annually associated with
more than 23,000 fatalities, more than 2 million injuries and cost over 56 billion Euros.

In economic terms this means that:

Around 70 billion Euro are spent each year on medical treatment of injured people in accidents
[4] and thousands man-years of work are lost. These numbers are bigger than the Gross Product
of several EU countries!
Social funds of magnitude also of billions Euro are devoted yearly to medical services and
rehabilitation for people, becoming temporarily or permanently disabled due to accidents.

A large proportion of all the costs are paid directly, or reimbursed, by insurance companies, which
therefore have a very significant interest in traffic safety enhancement.

In [9] two levels of monetary value of crashes are defined:

Economic values: include actual monetary loss, like medical care, legal services, vehicle
repair/replacement, lost productivity.
Comprehensive values: include both monetary loss and a valuation of less tangible human
consequences, such as pain and suffering and loss of life or disability.

Furthermore, the high monetary cost of the drowsy driver crash problem is presented. For heavy
vehicles the per-crash costs are even higher than these for passenger cars. The results are presented
below.

All vehicles Passenger Heavy

cars vehicles
Total annual US monetary cost
Economic values 3.8 billion 2.3 billion 312 million
Comprehensive value 12.5 billion 7.5 billion 852 million
Per-police-reported crash cost
Economic values 34,000 28,000 135,000
Comprehensive value 120,000 95,000 361,000
Crash cost per 100 million vehicles miles of travel
Economic values 169,000 156,000 345,000
Comprehensive value 570,000 510,000 938,000
Crash cost per registered vehicle annually
Economic values 20 20 180
Comprehensive value 68 60 490
Crash costs per vehicle produced over a full
operational life
Economic values 220 180 2,150
Comprehensive value 730 580 5,840
Table 1: Monetary estimates of the US drowsy driver crash problem in $ [9]

The benefit to the economy throughout Europe is therefore obvious, even if AWAKE manages
finally to prevent only a limited number of incidents, as, for example, a mere 5% reduction to road
casualties means about 2,500 less deaths and 75,000 less injuries per year, which sums up to around
500 MEuro gain annually for the European economy. Also since accidents due to loss of vigilance

8
tend to be more severe than the average, that figure may be even an underestimation of the actual
economic gain.

SAVE project has proved that such a system is feasible by integration and fusion of a wide rage of
sensors and that it works properly under restricted scenarios. AWAKE Consortium is determined to
develop it for actual highway applications.

REFERENCES

1. Petica, S., et. al., SAVE, D3.1, Driver needs and Public acceptance of Emergency Control
aids, August 1996.
2. Boussuge, J., Quinze ans de securite sur autoroute. Bilan et perspectives, Revue Gen. des
Routes et des Aerodromes, N. 726, 1995.
3. Garder, P., Continuous Shoulder Rumble-Strips A safety evaluation, In: CSRS Seminar,
Linkoeping, August 1998
4. Cerrelli, E.C., 1996, 1996 Traffic Crashes, Injuries and Fatalities preliminary report,
NHTSA DOIT HS 808 543.
5. FHWA, Crash problem Size Assessment: Large truck crashes related primarily to driver
fatigue, Federal Highway Administration, Office of Motor Carriers, September 1998
6. World health Organisation, World Health Statistics, Geneve, 1993
7. Knipling, R.R., Wang, J.S., Crashes and Fatalities related to driver drowsiness/fatigue,
Research Note, US Department of Transportation, 1997
8. CEESAR, Fatal traffic accidents main causes, Internal report for GIE PSA RENAULT, 1990.
9. Evans, L., 1991, Traffic Safety and the Driver, van Nostrand Reinhold (ed.), New York
10. Bekiaris, E., et. al., SAVE project Final Report, January 1999
11. NHTSA, 1998, Drowsy Driving and Automobile Crashes,
www.nhtsa.dot.gov/people/perform/human/ Drowsy.html
12. Wierwille, W.W., Lewin, M.G. and Fairbanks, R.J.III, 1996, September, Final report: Research
on vehicle-based driver status/performance monitoring, Parts I, II and III, Tech.Report No
DOT HS 88 638, Washington DC:NHTSA
13. NHTSA, 1998, Evaluation of Techniques for Ocular Measurement as an Index of fatigue and
the basis for alertness management, DOT HS 808 762
14. Renner, G., Mehring, S., 1997, Lane Departure and Drowsiness Two major accident causes
One Safety System, 4th World Congress on ITS, 21-24 October 1997, ICC Berlin, Germany
15. Nova-Assist Techn., Atlas Researches Ltd., 2000, Technical (Application) Notes
16. Boverie, S., et. al. SAVE, D8.1, Strategy for pre- and post-emergency actions, August 1997.
17. Bekiaris, E., et. al., SAVE, D4.1, Impact of SAVE system to Traffic Safety, December 1998.
18. National Transportation and Safety Board of US, Department of Transportation efforts in the
1990s to address operator fatigue, Safety Report NTSB/SR-99/01, May 1999
19. Horne, J. & Reyner, L, Vehicle accidents related to sleep - a review, Occupational and
Environmental Medicine, 1999, 56: 289-294
20. Stutts, J.C, Wilkins, J.W & Vaughn, B,V., Why do people have drowsy driving crashes? Input
from drivers who just did, AAA Foundation for traffic Safety, nov 1999, USA
21. Rau, P.S., NHTSA, A heavy Vehicle Drowsy Driver Detection and Warning System:
Scientific issues and Technical challenges, 1998
22. Moore-Ede, M., The Twenty-Four Hour Society, Addison-Wesley, 1993
23. Head Pose and Gaze Direction Trackings, www.seeingmachines.com/mid/rrtech.htm
24. SafeTRAC product for drowsiness detection, www.assistware.com