Survey on Stress detection while driving
Anoop Mathew
Student of DBIT
University of Mumbai
Email: [email protected]
Benjo Francis
Student of DBIT
University of Mumbai
Email: [email protected]
This paper provides a comparative analysis of five cases
of ways to detect stress while driving that have been widely
recognized as successful. Society will experience a rapidly
aging population over the next few decades, we still need to
maintain safety and security in our society. To successfully
achieve this goal, it is essential for us to predict and reduce
human errors and mishaps as far as is possible. In order
to reduce human errors, there have been many research pa-
pers published about various techniques to monitor stress or
fatigue while driving . This paper suggests factors associ-
ated with successful detection of stress , particularly in real
time driving and simulated driving environment. The paper
concludes with a discussion that the early detection of this
situation is very much necessary to improve awareness and
performance in drivers, which are important for road and
traffic safety we therefore suggest the best method to detect
stress at the earliest . We also took a survey from 16 drivers
, regarding the main reason for having stress while driving
, about 9 to 10 drivers suggest the stress is caused due to
the ongoing metro construction work , traffic and speeding 2
Wheeler’s .
1 Detailed Study of the papers
In one of the Paper, It presents a methods for collect-
ing and analyzing physiological data during real-world driv-
ing tasks to determine relative stress level among drivers.
Electrocardiogram, electromyogram, skin conductance",and
respiration were used. Analysis 1: Data during the rest",
highway, and city driving conditions is used to differenti-
ate three levels of driver stress with an accuracy of over 97
percentage across multiple drivers and driving days. Anal-
ysis 2: Compared continuous features, calculated at 1-s in-
tervals throughout the complete drive, with a metric of ob-
servable stressors created by independent coders from video-
tapes. Results show that, for most the drivers studied, the
skin conductivity and heart rate metrics are most closely cor-
related with driver stress level.
The subject wore five physiologiacal sensors , an elec-
trocardiogram on the chest , an electromyogram on the left
shoulder , a chest cavity expansion respiration sensor around
the diaphragm , and tweo skin conductivity sensores , one
on the left hand and one on the left foot . The sensors were
attached to a computer in the rear of the vehicle [1]
Fig. 1
In this paper they have focused on ECG monitoring, that can
now be performed with minimally wearable patches and sen-
sors.They have included three stress levels to determine the
state: low stress level, medium stress level and high stress
level . Using machine learning algorithms from the ECG
signals alone, we could achieve 88.24 percentage accuracy
in detecting the three classes of stress.[2]
In one of the paper , it investigaes drivers condition or be-
havior using Heart Rate and Breathing rate .The Heart Rate
and Breathing Rate signals are acquired with an adrenergic
sensor connected to a chest strap that is worn underneath the
drivers clothing. High level of stress compromises decision-
making skills of a driver, thus his performance and awareness
drastically decrease .As first step of our research, we present
the preliminary results we obtained using machine-learning-
based processing of publicly available HR and BR data , ac-
quired with wearablesensors during driving task . Experi-
ments have been conducted on a publicly available dataset of
driving simulated data. The classification problem is formu-
lated as a binary problem: stress vs no-stress.[3]
Fig. 2: In the above figure Comparison between the original signal
of a given subject (test set) and the distribution of the signals of the
rest of the subjects (training set). (a)HR of no-stress conditions. (b)
HR of stress conditions. (a) BR of no-stress conditions. (b) BR of
stress conditions
In this paper , Salivary amylase is used as a biomarker, as it
is considered to be one of theindicators of sympathetic ner-
vous activity. There were 20 healthy female subjects in their
early twenties were considered in this study. The time-course
change of their salivary amylase activity (sAMY) is analyzed
before and during the driving. In parallel, subjective evalua-
tion using a questionnaire and measurement of oculomotor (
motion of the eyes ) angle are conducted on each subject. A
psychological effect of motor-vehicle driving that could not
be easily detected by a subjective evaluation was rapidly and
quantitatively evaluated by a biomarker in saliva. Moreover",
the results suggested that operation of an non-driving-related
device might induce a significant reduction in the drivers ca-
pacity to concentrate on driving
Fig. 3: Driving units and measurement
Study 1: Firstly, three electrodes are placed on the subject
for the measurement of the oculomotor angle. Then, subjects
answer the questions identified in the questionnaire regard-
ing psychological state (Q1). Initially, each subject takes a
sitting position for 5 min in order to measure the individual
baseline (A 1-3 , baseline period, Fig.2). The minimum value
is set as the sAMY during the base line period (AMY base ).
Next, the subjects drive for 21 min using the simulator, and
sAMY is measured every 3 min (7 times) . The oculomo-
tor angle is measured continuously during the driving period.
Immediately after completion of the measurement, the sub-
ject fills in the psychological state questionnaire (Q2). The
subjects are instructed to drive a car in the center of the lane
as much as they possibly can. Study 2: In order to investi-
gate the effects of operation of a device, which is not directly
related to driving a motor-vehicle, the subjects are instructed
to input digit on the touch panel of the car navigation device",
which is installed in the driving unit (non-driving-related de-
vice). With a method similar to Study 1, the subjects simul-
taneously perform two tasks, driving and inputting a number
on the car navigation device and sAMY and other parameters
are measured.[4]
In this paper, they intends to understand stress level detection
of a driver during real world driving experiment. This detec-
tion is based on heart rate variability (HRV) analysis which
is derived from ECG signal and reflects autonomic nervous
system state of the human body. In our study, the ECG sig-
nal of the driver is extracted and pre-processed in order to
perform the HRV analysis.
This analysis is accomplished us-ing one of the domain anal-
ysis approach such as time, frequency, time-frequency or
non-linear methods including Wavelet and STFT. After HRV
analysis, several parameters are extracted to build a vector
of features for the classification phase. Our system archi-
tecture , It includes three main phases including data anal-
ysis, feature Extraction and classification.The first step in
our analysis phase is the preprocessing which includes the
filtering of raw ECG Signal from all noise sources using a
bandpass filter The second step is detecting the R peak which
indicates the most visible and highest peak of the ECG sig-
nal. Therefore the filtered signal is passed through different
phases such as derivative, squaring and window integration.
In the last step of the analysis phase it is made to derive the
heart rate variability (HRV) which is the beat-to-beat alter-
ations in heart rate (HR). Under the resting conditions of the
testers, ECG of healthy individuals exhibits periodic varia-
tion in R-R intervals
Out of all Classifiers , Using SVM-RBF classifer stress de-
tection could be predicted with an accuracy of 83% . This
also points out therobust-ness of ECG biometric as an accu-
rate physiological indicator of a driver state. Results suggest
that the QRS complex are dominant in any case regardless
of each driver’s physiological driving conditions experienced
by the automotive drivers and can be used as an accurate
non invasive biometric feature without using the whole ECG
morphology. In future work, it is suggested to differentiate
the driver stress into three different levels low, medium and
high level using more samples and identifying significant pa-
rameters .[5]
2 Analysis Review
From the 5 papers that we compared on the basis of their
performance",efficiency",technology used, we suggest that the
method suggested in the paper Machine Learning for Stress
Detection from ECG Signals in Automobile Drivers by Ke-
shan, Parimi and Bichindaritz[2] is more efficient compared
to other papers. In this paper they use modern method
like machine learning for detecting stress. They also ob-
tain their results by using minimum number of wearable
sensors. The result they obtain for stress detection while
driving are in 3 levels;low, medium, high. According the
paper we did study on, they got an efficiency of 100%
for low and high level stress.And 88.24% accuracy for
medium level stress.The ECG signals of drivers stress used
in this study were obtained from MIT-BIH PhysioNet Multi-
parameter Database.The data from 17 participating drivers
were taken and the database has eight types of raw data
time stamp, ECG, electromyogram(EMG), foot galvanic
skin response(GSR), hand GSR, intermittent heart rate(IHR)",
marker, and respiration. The ECG signals of drivers that
were used in the system for detecting stress were obtained
from MIT-BIH PhysioNet Multi-parameter Database. The
database consisted of 8 set of data; ECG, time stamp",
intermittent heart rate (IHR), foot galvanic skin response
(GSR)",hand GSR, marker, and respiration, electromyogram
(EMG). For feature extraction they have extracted 14 differ-
ent features for analysis which include: Average Difference
Beats, Average Beats, Average QR interval and more. They
used WEKA as a classification tool to classify these extracted
features. It consists of many classifiers offered by weka",
out of which 10 algorithms from varied types were selected
for classification to perform their comparative study, namely",
IBK (K-nearest- neighbors), Nave Bayes, RandomForest",
Multilayer Perceptron, SMO (support-vector machine), Lo-
gistic Regression, ZeroR, J48 (decision tree), IB1 (1-nearest
neighbor) and RandomTree. In this paper even though they
tried using different methods to get maximum efficiency",
they obtained the maximum efficiency using Naive Bayes
algorithm. It gave 100% efficience for low and high stress
levels. Where as in other algorithms like J48 with 98% effi-
cency for the above cases.
3 Conclusion
In this paper an analysis of five papers related to stress de-
tection while driving is proposed .This paper shows the ef-
ficency of each stress detection technique .[1] In the future",
we may want vehicles to be more intelligent and responsive",
managing information delivery in the context of the drivers
situation. Physiological sensing is one method of accom-
plishing this goal. This study tested the applicability of phys-
iological sensing for determining a drivers overall stress level
in a real environment . The results showed that there could
be three stress levels that could be recognized with an over-
all accuracy of 97.4 percentage .[2] Using machine learning
algorithms from the ECG signals alone, we could get 88.24
percentage accuracy in detecting the three classes of stress:
low, medium and high. Also, we have obtained a perfect
100 percentage accuracy in detecting high stress levels with
NaiveBayes.[3] Results demonstrated that both HR and BR
can be used for stress prediction and that their combination
is more robust than each signal considered separately. The
accuracy reached by the best combination is about 70 per-
centage We demonstrated that the accuracy level which can
be obtained using only HR and BR sensors is very interest-
ing, therefore they can be considered very promising for the
integration into a machine-learning-based driver monitoring
system. [4] Evaluation of driver stress using a biomarker is
considered to be very useful to improve the safety and se-
curity of motor-vehicle drivers by quantification of driving-
induced stress.Moreover, the results suggested that operation
of an non-driving-related device might induce a significant
reduction in the drivers capacity to concentrate on driving
.[5] ECG signal converted into HRV for analysis reflects the
response of autonomic nervous system and allows to predict
stress level of the driver which will ensure to provide better
security for automotive drivers in order to avoid car thefts
and to recognize and identify the automotive drivers in dif-
ferent physiological driving conditions . SVM with a RBF
kernel presented the best results giving the highest rates of
correct prediction .In future work, the paper is suggested to
differentiate driver’s stress level into three different levels
low, medium and high level using more samples and iden-
tifying significant parameters to realize the purpose.
4 References
[1] Jennifer A. Healey and Rosalind W. Picard, ”Detecting
Stress During Real-World Driving Tasks Using Physiologi-
cal Sensors ”",IEEE Transactions on Intelligent Transporta-
tion System, Vol. 6, No. 2, 156 - 166, June 2005
[2] I. Bichindaritz",N. Keshan, and P. V. Parimi ",”Machine
Learning for Stress Detection from ECG Signals in Automo-
bile Drivers ”",2015 IEEE International Conference on Big
Data (Big Data) ",29 Oct.-1 Nov. 2015
[3] Paolo Napoletano and Stefano Rossi ",”Combining heart
and breathing rate for car driver stress recognition ”, Interna-
tional Conference on Consumer Electronics - Berlin (ICCE -
Berlin)
[4] M. Yamaguchi, J. Wakasugi, J. Sakakima ",”Evaluation
of Driver Stress Using Biomarker in Motor-vehicle Driving
Simulator ”, proceedings of the 28th IEEE EMBS Annual
International Conference New York City",Aug 30-Sept 3.
[5] Nermine Munla, Mohamad Khalil, Ahmad Shahin and
Azzam Mourad ",” Driver Stress Level Detection Using HRV
Analysis ”",2015 International Conference on Advances in
Biomedical Engineering (ICABME), 16-18 Sept. 2015.
Detailed Study of the papers
Analysis Review
Conclusion