Sunteți pe pagina 1din 4

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251286028

Predicting Accident Probability from Frequency of Microsleep Events

Article · January 2009


DOI: 10.1007/978-3-642-03882-2_607

CITATIONS READS

5 99

5 authors, including:

Bill Sirois Udo Trutschel


Circadian Institute for System Analysis and Applied Numerics
18 PUBLICATIONS   72 CITATIONS    133 PUBLICATIONS   1,040 CITATIONS   

SEE PROFILE SEE PROFILE

David J Edwards David Sommer


Caterpillar Inc. SDA GmbH Schmalkalden
26 PUBLICATIONS   348 CITATIONS    101 PUBLICATIONS   429 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Driver Sleepiness View project

Alertness management View project

All content following this page was uploaded by Martin Golz on 12 April 2016.

The user has requested enhancement of the downloaded file.


Predicting Accident Probability from Frequency of Microsleep Events
B. Sirois1 , U. Trutschel1,4 , D. Edwards2 , D. Sommer3 and M. Golz3,4
1 Circadian Technologies Inc., Stoneham, Massachusetts, USA
2 Machine Research, Caterpillar Inc., Peoria, Illinois, USA
3 Faculty of Computer Science, University of Applied Sciences Schmalkalden, Germany
4 Institute for System Analysis and Applied Numerics, ISAAN, Tabarz, Germany

Abstract— There is growing acknowledgement that fatigue a problem in all 24-hour operations, but the real danger is
plays a significant role in traffic accidents. This study evaluated the occurrence of MSE. During MSE, the operator is inca-
the relationship between fatigue-related accidents and Micro- pable of responding to outside events. The brain has, in some
Sleep Events (MSE) using a real car driving simulator. Ten sense, turned its self off for a few moments. If nothing in the
healthy volunteers participated in the study for two nights each. work environment changes during a MSE, there are no con-
After a period of struggling against sleep pressure, all subjects sequences. But if something happens, if there is a curve in the
experienced periods of extreme fatigue and Micro-Sleep Events road, if the car in front breaks, or your truck drifts of out its
(MSE). Most MSE had no consequences, but some resulted di- lane, there is an immediate risk of an accident. Perhaps the
rectly in an accident. To establish a correlation between MSE most frightening thing about MSE is that they mostly happen
and fatigue related accidents all MSE were counted separately without awareness or forewarning. This has been observed
for driving sessions with and without accidents. Accident proba- numerous times during our night time driving simulator stud-
bility was calculated as the ratio of the number of MSE in the ac- ies.
cident sessions to the total number of MSE. Under the extremely Despite the high accident potential of MSE, most research
fatiguing protocols the overall accident probability was 35 %. in this field was and is focused on general driver fatigue
[Torsval et al. 1987], [Jung et al. 1997].
Keywords— Accident probability, Micro Sleep Event Duration, Recently, the MSE detection issue received some long
Correlations Micro-Sleep Events - Accident Events
overdue attention [Blaivas et al. 2007], [Davidson et al.
2007]. However, the direct quantitative relationships between
I. I NTRODUCTION MSE frequency and accident probability have not yet been
reported. To establish such a relationship requires a reliable,
In general, driver fatigue [Mackie 1977], [Moore-Ede continuous MSE detection capability.
1993] can be characterized by many features, such as a de- However, many researchers try to explain MSE based on
crease in mental and physical performance, reduced vigi- certain EEG rules, similar to the sleep onset classification.
lance, loss of cognitive and logical reasoning skills, impaired [Sommer et al. 2001] have shown that the EEG characteris-
judgment, reduced motor coordination, slower reaction time, tics during MSE is quite complex and that only rule free clas-
absentmindedness, mental lapses, uncharacteristic mistakes, sification tools allow reliable MSE detection based on EEG.
erratic behavior, and the occurrence of Micro-Sleep Events In [Golz et al. 2007] and [Davidson et al. 2007], MSE classi-
(MSE). It is important to note that many individuals are either fication based on EEG and EOG analysis utilizing rule free,
not aware of the degree of impairment produced by fatigue, modern classification tools, such as Support Vector Machines
or they are not able to recognize the early symptoms of MSE. and Tapped delay-line Multilayer Perceptron networks, were
Within the public sector, inattention and driver fatigue (in- presented. [Paul et al., 2005] have shown that an increasing
cluding MSE) are significant factors in commercial truck number of MSE can be associated with worsening overall
crashes, conservatively accounting for 1,200 deaths and driving performance.
76,000 injuries annually at an estimated cost of $12.4 billion.
Fatigued drivers are often not aware of their condition, fre-
quently driving for repetitive episodes of 3-12 seconds with II. E XPERIMENTAL S ETUP
their eyes totally closed. These drivers are experiencing dan-
The study was conducted in the laboratory of adaptive
gerous Micro-Sleep Events (MSE) because they are mentally
biosignal analysis at the University of Applied Sciences,
incapacitated during the MSE periods. In some accident re-
Schmalkalden (Germany). The lab has a sophisticated, real
ports fatigue is named as contributing factor, when the real
car driving simulator and extensive experience conducting
factor was very likely a single MSE. Of course, fatigue is
these types of studies. The equipment used is fully controlled
by specialized software, allowing regulation of light levels, ously observed behavior. The AE probability for a single time
car environment (temperature, noise level, humidity), and of window was defined on a set of binary values, where ’zero’
controls and instruments. Ten volunteers participated in the means that no AE occurred during the whole 4 minute time
study, 8 men and 2 women. The average age was 25.2 ± 3.5 window and ’one’ means that at least one or more AE oc-
(range 20 - 29). They were all healthy, and kept regular sleep curred in the 4 minute window. The overall AE probability
/ wake schedules. Volunteers arrived at the lab at 11 pm. Af- was thus defined as the ratio of the number of time windows
ter wire-up, experimental sessions started at 1 am. There were with at least one AE and the total number of time windows.
seven experimental sessions, with the last session starting at 7
am. Each driving session had duration of 40 minutes. Several IV. R ESULTS
bio-signals were recorded: EEG (FP1, FP2, C3, Cz, C4, O1,
O2, A1, A2), EOG (vertical, horizontal), ECG, and EMG. As expected, the percentage MSE duration and AE proba-
In addition, three video recordings (driver’s head & pose, bility increased progressively through the night. This can be
driver’s eyes, driving scene) were stored. Extreme values of observed on a subject specific level (Figure 2, 3) and for all
Variation of Lane Deviation (VLD) were used to determine 10 subjects (Figure 4) on average. As a general observation,
the Accident Events (AE). MSE and AE scoring began im- we found that the AE probability moved from 0 to 1 as soon
mediately and continued throughout the driving sessions by as the MSE duration reaches 0.5 (50 % of the 4 minute time
two researchers who watched the video streams. Time stamps window). This can be considered as an empirical threshold.
were set for typical signs of MSE (e.g. prolonged eyelid clo-
sures, rolling eye movements, head nodding) and major AE
(e.g. driving incidents and drift-out-of-lane accidents). In all,
2,290 MSE were found (per subject: mean number 229 ±
67, range 138 - 363) and 668 AE (per subject: mean number
66.8 ± 54, range 7 - 193). Typical sequences of MSE and AE
are depicted in figure 1.

Fig. 2: Percent MSE duration and AE probability for individual subject A


(Definitions of MSE duration [%] and AE probability are based on 4 minute
time window). Please note that the AE are occurring at times when MSE
duration exceeds 50 % in a given 4 minute time window.
Fig. 1: Scored MSE (red dots) and serious AE (blue bars).
As shown in Figure 3, the average MSE durations were
MSE occur first, sometimes leading to an AE, sometimes not larger than 0.6 (60 % of a 4 minute time window) of
not. From visual observation, there seemed to be a strong cor- the whole driving session. Descriptive statistics of the results
relation between the frequency of MSE per time interval and of consecutive classification clearly confirmed time-on-task
the AE probability. and time of day effect regarding MSE and AE occurrences.
Within and between driving sessions average probability of
MSE (red bars) and average probability of AE increased. It
III. DATA A NALYSIS
should be noted that there was a slight decrease of MSE du-
Based on these visual observations of many MSE and AE, ration from 6th to 7th driving session, but AE probability in-
a moving time window of 4 minutes was selected. The rel- creased further.
ative MSE duration in percentage was then calculated based These results clearly reflect the high correlation coeffi-
on this time window. For example, if the total length of all cients in the later driving sessions. There is a clear relation-
MSE was 2 minutes inside the 4 minute time window, then ship; longer MSE durations result in higher AE probability.
the MSE duration would 50% (Sessions 4-7 in Figure 2, red In general, for MSE durations smaller than 10 %, accidents
bars). The consecutive detection of MSE from EEG and EOG can happen, but the probabilty is low (Session 1 in Figure 4).
was performed by Support Vector Machines [Sommer et al., For MSE durations above 50 %, accidents will occur with a
2008]. This methodology enables not only the detection of probability close to 100 %, clearly exceeding the overall AE
clear MSE scored by the raters, but also MSE with ambigu- probability of 35 %.
View publication stats

ACKNOWLEDGEMENTS
Part of this work was supported by the German Federal
Ministry of Education and Research through the ”Applica-
tive Research and Development at Universities of Applied
Sciences” program.

R EFERENCES
Fig. 3: Percent MSE duration and AE probability for individual subjects B 1. Blaivas AJ, Patel R, Hom D, Antigua K, Ashtyani H., (2007). Quantify-
(Definitions of MSE duration [%] and AE probability are based on 4 minute ing microsleep to help assess subjective sleepiness, Sleep Med. 2007 Mar;
time window). Please note that the AE are occurring at times when MSE 8 (2),156-9.
duration exceeds 50 % in a given 4 minute time window. 2. Davidson, P.R., Jones, R.D., Peiris, M., (2007). EEG-Based Lapse De-
tection with High Temporal Resolution, IEEE TRANSACTIONS ON
BIOMEDICAL ENGINEERING, VOL. 54, 832-839.
3. Golz M, Sommer D, Chen M, Trutschel U, Mandic D (2007). Feature
V. C ONCLUSIONS Fusion for the Detection of Microsleep Events. J VLSI Signal Proc Syst
49: 329-342.
The results clearly demonstrated the close quantitative re- 4. Jung, T-P, Makeig, S, Stensmo, M, Sejnowski, T. (1997). Estimating of
lationship between MSE and AE probability. As one conse- alertness from EEG power spectrum, IEEE Transactions on Biomedical
Engineering 44, pp. 60-69.
quence of the presented results, there should be a redefinition 5. Mackie, R. (1977). Vigilance: Theoriy, Operational Performance, and
of driver fatigue involving MSE. The concept of MSE goes Physiological Correlates, Plenum Press, New York, London.
beyond a simple driver fatigue phenomena, because it is im- 6. Moore-Ede, M. (1993). The Twenty Four Hour Society: Understanding
The Human Limits In A World That Never Stops” Addison-Wesley.
plying the loss of control during driving. MSE are a frequent 7. Paul, A., Boyle, L., Tippin, L., Rizzo, M. (2005). Variability of Driving
transitional phenomenon occurring in normal subjects who Performance During Microsleeps, Proceedings of the Third International
are engaged in a monotonous, continuous task for extended Driving Symposium on Human Factors in Driver Assessment, Training
and Vehicle Design.
periods of time. This is of particular importance to the trans- 8. Sommer, D., Golz, M.; Trutschel, U., Ramsthaler, C.; Moore-Ede, M.
port and mining sectors, where there is a need to maintain (2001). Characterization of the Electroencephalogram of Microsleep using
sustained attention for extended periods of time, and where Self-Organized Feature Maps Proceedings on the Computational Intelli-
gence: Methods and Applications (CIMA’2001), Kuncheva, L.; Steinman,
MSE can lead to multiple-fatality accidents and high value F.; Haefke, C.; Aladjem, M.; Novak, V. (eds.), pp 123-129 Bangor, Wales,
property damage. The presented results provide a better un- UK.
derstanding of which MSE levels represent a critical thresh- 9. Sommer, D., Golz, M. & Krajewski, J. (2008). Consecutive detection of
driver’s microsleep events. In J. Vander Sloten, P. Verdonck, M. Nyssen, J.
old that, when exceeded, will inevitably produce accidents. Haueisen (Eds.), IFMBE Proceedings, 22, (pp. 243-247). Springer: Berlin.
10. Torsvall, L, Akerstedt, T, (1987). Sleepiness on the job: continuously
measured EEG in train drivers, Electroencephalography and Clinical Neu-
rophysiology 66 , pp.502-511.

Author: Udo Trutschel


Institute: Circadian Technologies Inc.
Street: 2 Main Street, Suite 310
City: Stoneham
Country: USA
Email: utrutschel@circadian.com

Fig. 4: Correlation between percent MSE duration and AE probability


(Definition of MSE duration and AE probability is based on 4 minute time
windows). Please note that the correlations are higher with increasing MSE
durations.

S-ar putea să vă placă și