P.L.P. Rau (Ed.): Internationalization, Design, HCII 2011, LNCS 6775, pp. 258–264, 2011.
© Springer-Verlag Berlin Heidelberg 2011
Driving Distraction Analysis by ECG Signals:
An Entropy Analysis
Lu Yu, Xianghong Sun, and Kan Zhang
Institute of psychology, Chinese Academy of Sciences, 4A, Datun Doad,
Chaoyang District, Beijing, China, 100101
Abstract. This paper presents a novel method in driving distraction analysis:
entropy analysis of ECG signals. ECG signals were recorded continuously while
15 drivers were driving with a simulator. Mental computation task was employed
as driving distraction. Sample entropy and power spectrum entropy of drivers.
ECG signals while they were driving with and without distraction were derived.
The result indicated that entropy of drivers ECG signals was sensitive to driving
distraction and were potential significant metrics in driving distraction
Keywords: Entropy, Driving distraction, ECG signal.
The increasing use of on-board electronics and in-vehicle information systems has
made driving distraction a major concern in the driving safety field . When drivers
manage another task while driving, e.g. listening to the radio, holding a cell-phone
conversation, employing on-board navigation system, the distraction of attention will
decrease their performance, even causes traffic accident.
The analysis and recognition of driving distraction is significant to safety since it is
the basis of avoiding distraction and the design of in-vehicle information system (IVIS)
and other driving aided devices. Most of the studies focused on driving behavior and
performance related to distraction. Eye movement and driving performance were
investigated to analyze and identify the distraction in driving . D’Orazioa et.al.
established a visual framework to estimate the drivers’ inattention while driving with
secondary task . Reaction Time on secondary task along with driving performance
was recorded to explore driving distraction activity also .
Model for inferring psychological significance from physiological signals has been
built since 20 years ago . Physiological signals have been widely used in emotion
recognition    . However, to the limit of our knowledge, only a few study
investigated drivers’ functional state associated with driving workload by physiological
indices. Jennifer, Healey and Rosalind monitored drivers’ physiologic reactions during
real-world driving situations using physiological sensors . Electrocardiogram
(ECG), Electromyogram (EMG), Electrodermal activation (EDA) and respiration were
Driving Distraction Analysis by ECG Signals 259
recorded and were used to evaluate the stress of drivers in different driving task. Collet,
Clarion and Morel et.al. evaluated the strain undergone by drivers when they managed
the secondary task while driving. Electrodermal activity and instantaneous Heart Rate
(HR) were recorded . ECG is one of the widely used tools to explore cognitive
requirements of complex task performance . Most of the studies employed
Heart Rate (HR) and Heart Rate Variability (HRV) to evaluate mental workload
. Few studies had explored the correlation between the original ECG signal and
In the present research, we explored drivers’ distraction by the original ECG signals
through entropy analysis. In our experiments, drivers drove with and without secondary
task respectively in a driving simulator. ECG signals were recorded while operation
and performance data were recorded either. The ECG signals were analyzed in time
domain and spectrum domain. Sample entropy can describe the complexity of a time
series. Our hypothesis was that the complexity of the ECG signals of drivers with and
without distraction would be different significantly. Therefore, we calculated the
sample entropy in different time scales of the original ECG signals and compared the
values in the two situations. Power spectrum entropy is often used in biomedical
engineering, e.g. cerebral ischemia detection , sleep stages , myocardial
infarction patients diagnosis . In the present research, we derived the power
spectrum entropy of the original ECG signals and tried to make it one of the metrics of
drivers’ distraction. The result indicated that the sample entropy and the power
spectrum entropy of the original ECG signal were sensitive to driving distraction.
2 Material and Methods
The participants were 15 licensed drivers aged from 18 to 50 years (mean 25, SD 5.2).
There were 7 males and 8 females. All the participants were healthy and not receiving
medication. They gave their informed consent after having been informed about the
main contents of the experiment and were paid for their participation.
The experiment took place in a driving simulator. The driving environment was a
one-way driving, three lanes highway scene in the simulated driving task. The driving
task was car following. Drivers were required to follow the head car while it changed
lanes, stepped on the accelerator or brake. They should operate the steering wheel,
accelerator or brake of the simulator so that the following car can close to the head car
as near as possible while avoiding collision. The management of the dual-task was
made under driving conditions. The secondary task was double-digital addition mental
computation. After 20 minutes exercises which made drivers familiar to the operation
of driving simulator, the drivers performed the experiment included two sessions:
driving without secondary task and driving with secondary task. The mental
260 L. Yu, X. Sun, and K. Zhang
computation problems were presented to the drivers by a clear female voice through
earphones while they were driving. Each problem was presented for 10 seconds.
Participants spoke the result of their computation to the microphone. The number of
mental computation problems was 60. Each session lasted 10 minutes.
The driving simulator consisted of two parts: first, simulated car operation device
including control stick, accelerator and brake pedal. Second, driving behavior
surveillance system based on computer consisted of driving task presentation,
recording of driver’s reaction, data management, feedback of driving state and
alarming modules. ECG signal was recorded by KF2 dynamic multi-parametric
physiological detector. This kind of wireless wearable physiological detector can
record multiple physiological indices include ECG with 3 leads, respiration and body
temperature. The data can be analyzed by a data processing software. Figure 1 shows
the physiological detector and the driving simulator.
Fig. 1. The KF2 dynamic multi-parametric physiological detector and driving simulator
Driving Distraction Analysis by ECG Signals 261
2.4 Data Collection
Driving performance and ECG data were collected. The driving performance data was
recorded by the computer integrated with the simulator. Participants wore the
physiological detector before the start of the experiments and began to record ECG signal
until the end of the experiments. The ECG signal was sampled at the rate of 250 Hz.
3 Data Analysis
Physiological parameters were related to the Autonomic Nervous System (ANS)
functioning. The ANS is known to give a close estimation of subjects’ arousal
especially through the orthosympathetic branch  specialized in mobilizing energy
resources in response to internal and external milieu demands . The energy
resources needed in driving with and without secondary task should be different. Thus
the physiological features should be different in the two situations. We derived sample
entropy and power spectrum entropy of the original ECG signals in the two cases and
tried to explore the effect of driving distraction in ECG signal. Original ECG signals
were first denoised by discrete wavelet coif4. Sample entropy and power spectrum
entropy were calculated consequently.
Sample Entropy is a statistic representing the self similarity of a time series . The
more complex the time series is, the larger the sample entropy is. In other words, the
more self-similar the time series is, the fewer the sample entropy is. Fewer data is
needed to derive robust estimation of sample entropy compared to some other statistics
such as approximate entropy, kolmogorov entropy. Thus sample entropy is widely used
in the study of experimental clinical cardiovascular and other biological time series.
The calculation of sample entropy can be seen in . Furthermore, on account of the
multiple time scales inherent in healthy physiologic dynamics, Costa et.al. introduced
multi-scale sample entropy . We calculated the multi-scale entropy of the ECG
signals in the two experiment sessions respectively. The scales lasted from 1 to 10.
Power Spectrum Entropy (PSE) is the entropy of the power spectrum of a time
series. It describes uncertainty of the energy distribution of the time series in each
frequency. The larger the PSE is, the more uniform the energy distribution is. The
power spectrum entropy of the ECG signals in 6 frequency bands which corresponded
to the main components of ECG signal were calculated.
First, the multi-scale entropy of drivers with and without distraction are significant
different. The univariate repeated measures F-test of the multi-scale entropy showed
that significant levels in all the scales are less than 0.05 except for scale 1
(F(1,14)=3.641, p=0.077). Figure 2 describes the difference between the two cases.
Sample entropy with distraction in each scale was larger than that without distraction
which indicated that distraction would increase the complexity of ECG signal.
262 L. Yu, X. Sun, and K. Zhang
Table 1 is the PSE of the drivers’ ECG signals with and without distraction. Line 4
and 5 are the result of univariate repeated measure which indicated the significant effect
of distraction in the PSE of the critical bands of ECG signal.
Fig. 2. Multi-scale entropy of drivers’ ECG signal with and without distraction
Table 1. Power Spectrum Entropy of drivers’ ECG signals
Whole band 0~1.5Hz 0~4Hz 0~8Hz 0~20Hz
4.0600(0.6283) 0.2332(0.1423) 0.5677(0.2934) 1.5585(0.7053) 3.5558(0.6945)
4.1270(0.5965) 0.3169(0.2432) 0.6455(0.3528) 1.6213(0.7164) 3.6239(0.8727)
F(1,14)=4.5 F(1,14)=3.592 F(1,14)=3.913 F(1,14)=5.142 F(1,14)=5.694
p 0.052 0.079 0.068 0.040 0.032
The value of PSE in table 1 is mean and standard deviation (in the bracket) .
Entropy analysis is a non-linear dynamic method. ECG signal is a kind of complex
non-linear signal. Result of our experiments indicated that entropy analysis of ECG
signal was meaningful to the analysis of drivers’ functional state in driving distraction.
Sample entropy can describe the complexity of a time series. Figure 2 showed that the
sample entropy with distraction was larger than that without distraction. We tried to
Driving Distraction Analysis by ECG Signals 263
give a possible explanation that the distraction increased the mental workload which
changed the functional state of the drivers. The change increased the complexity of the
ECG signals. Similarly, we found that the power spectrum entropy of ECG signals
which described the uncertainty of the energy distraction in spectrum domain was
different in the two situations. The PSE in driving with distraction was larger than that
in driving without distraction also. The results of the entropy in time domain and in
spectrum domain are consistent.
The results of our experiment indicate that entropy analysis of ECG signals in driving is
meaningful. The significant difference of entropy of driving with and without
distraction shows that entropy of ECG signals is sensitive to driving distraction. This
make entropy of ECG signals, either in time domain or in spectrum domain, be
potential significant metrics in driving distraction measurement. Consequently, this
implies that we can get benefit from the entropy of ECG signals in the recognition of
driving distraction which is significant in some engineering psychology studies.
Acknowledgment. This work was supported by the 45''th Chinese postdocoral science
1. Lianeras, R.: NHTSA driver distraction Internet forum: Summary and proceedings (On-line
conference proceedings). National Highway Traffic Safety Administration, Washington (2000),
2. Liang, Y., Reyes, M.L., Lee, J.D.: Real-Time Detection of Driver Cognitive Distraction
Using Support Vector Machines. IEEE Transactions on Intelligent Transportation
Systems 8(2), 340–350 (2007)
3. Zhang, Y., Owechko, Y., Zhang, J.: Learning-Based Driver Workload Estimation. In:
Prokhorov, D. (ed.) Comput. Intel. in Automotive Applications. SCI, vol. 132, pp. 1–24.
Springer, Heidelberg (2008)
4. D’Orazio, T., Leo, M., Guaragnella, C., Distante, A.: A visual approach for driver
inattention detection. Pattern Recognition 40, 2341–2355 (2007)
5. Tango, F., Botta, M.: Evaluation of Distraction in a Driver-Vehicle-Environment
Framework: An Application of Different Data-Mining Techniques. In: Perner, P. (ed.)
ICDM 2009. LNCS, vol. 5633, pp. 176–190. Springer, Heidelberg (2009)
6. Carcioppo, C.J., Tassinary, L.G.: Inferring psychological significance from physiological
signals. American Psychologist 45(1), 16–28 (1990)
7. Picard, R.W., Vyzas, E., Healey, J.T.: Machine Emotional Intelligence: Analysis of
affective physiological state. IEEE Transactions on Pattern Analysis and Machine
Intelligence 23(10), 1175–1191 (2001)
8. Ekman, P.: An argument for basic emotion. Cognition and Emotion 1992(6), 169–200
264 L. Yu, X. Sun, and K. Zhang Download full-text
9. Stemmler, G., Heldmann, M., Pauls, C.A.: Constrains for emotion specificity in fear and
anger: the context counts. Psyehophysiology 38, 275–291 (2001)
10. Kim, K.H., Bang, S.W., Kim, S.R.: Emotion recognition system using short-term
monitoring of Physiological signals. Med. Biol. Eng. ComPut. 42, 419–442 (2004)
11. Jennifer, A.H., Rosalind, W.P.: Detecting Stress During Real-World Driving Tasks Using
Physiological Sensors. IEEE Transactions on Intelligent Transportation Systems 6(2),
12. Collet, C., Clarion, A., Morel, M., Chapon, A., Petit, C.: Physiological and behavioral
changes associated to the management of secondary tasks while driving. Applied
Ergonomics 40, 1041–1046 (2009)
13. Hankins, T.C., Wilson, G.F.: A comparison of heart rate, eye activity, EEG and subjective
measures of pilot mental workload during flight. Aviation, Space, and Environmental
Medicine 69, 360–367 (1998)
14. Wilson, G.F.: Air-to-ground training missions: A psychophysiological workload analysis.
Ergonomics 36, 1071–1087 (1993)
15. Wu, H.J., Zhang, H., Zheng, C.X.: Application of Spectrum Entropy to the Noninvasive
detection of focal ischemiccerebal Injury. Journal of Biomedical Engineering 20(2),
16. Liu, H., He, W., Chen, X.: Nonlinear Dynamic Method of Sleeping EEG. Journal of Jiangsu
University 26(2), 174–177 (2005)
17. Chang, J.: Application of Entropy Analysis in Bioinformation Processing. Master Thesis,
Southwest University (2008)
18. Boucsein, W.: Psychophysiology in the work place – goals and methods. In: Ullsperger, P.
(ed.) Psychophysiology of Mental Workload, pp. 35–41. Bundesanstalt fur Arbeitmedizin,
19. Wallin, B.G., Fagius, J.: The sympathetic nervous system in man: aspects derived from
microelectrode recordings. Trends Neuroscience 9, 63–67 (1986)
20. Richman, J.S., Moorman, J.R.: Physiological Time Series Analysis using Approximate
Entropy and Sample Entropy. American Journal of Physiology - Heart and Circulatory
Physiology 278(6), H2039–H2049 (2000)
21. Costa, M., Goldberger, A.L., Peng, C.K.: Multiscale Entropy Analysis of Complex
Physiologic Time Series. Physical Review Letters 89(6), 068102-1-1–068102-1-4 (2002)