2017 International Conference on Intelligent Computing and Control (I2C2)

Drunk Driving and Drowsiness Detection

Vivek R. Nair Nadir N. Charniya
Department of EXTC, VESIT Prof. Department of EXTC, VESIT
Mumbai, India Mumbai, India
[email protected] [email protected]

Abstract—Development of safety features to prevent drunk Machine (SVM) and Logistic Regression Models (LRM) to
and drowsy driving is one of the major technical challenges in the recognise driver distraction. Thus, making it convenient for
automobile industry. Driving while being drunk or drowsy is a signal acquisition but highly dependent on the type of vehicle,
major reason behind road accidents especially in the modern age. driver experience, and the road conditions [2].
Driving when drowsy can lead to higher crash risk than being in
alert state. Therefore, by using assistive systems to monitor The feature-based approach analyses visual features from
driver’s level of alertness can be of significant help in prevention the driver’s facial images having unique features such as eye
of accidents. This paper aims towards the detection of driver’s blinks, yawning, and head movements. Hammoud et al.
drowsiness using the visual features approach along with drunk proposed a drowsiness detection system that would assess the
detection using alcohol sensor. Driver drowsiness is based on real- eye status in the near-infrared spectrum. Moriyama et al.
time detection of the driver’s head, face and mouth, where-in predicted the eye state by using templates to match with the
HAAR-Cascade classifier for face and eye detection and template
matching in the mouth region for yawning detection. The system
eyelids. Percentage of eyelid Closure (PC) which counts the
will also have an alcohol detection sensor which will determine number of eye blinks of the driver is another widely used
whether the driver is drunk or not, thus covering the major technique for drowsiness detection. For practical applications,
reasons behind road accidents. visual-feature-based approaches are preferred since they are
instinctive and does not bother the driver [2].
Keywords—drowsy, drunk, visual features, HAAR-Cascade
classifier, driver safety, alcohol sensor.

I. INTRODUCTION
Driving while drowsy or in drunk state, are the two main
reasons for traffic accidents and the related financial losses.
Researchers have been working on designing driver drowsy
monitoring systems over the last decade. Though, there have
been numerous improvements in driver safety, yet a significant
number of serious accidents still occur all over the world. The
U.S. National Highway Traffic Safety Administration each year
reports roughly about 60,000 traffic accidents taking place due
to sleepiness related problems. studies indicate that 25%–30%
of driving accidents are caused due to drowsiness [1]. In 2008,
the National Highway Traffic Safety Administration estimated
about 100,000 police reports on automobile crashes which were
direct results of driver being drowsy resulting in over 71 000
injuries, 1550 deaths, and $12.5 billion in financial losses [2].
To detect driver inattention, three main approaches have been
developed i.e. via, Physiological, Driving-behaviour-based, and
Visual-feature-based approaches. Physiological methods
involve analysing of vital signals such as brain activity, heart
rate, and pulse rate. However, this technique requires use of
electrodes to be attached to the driver’s body, which is intrusive
in nature therefore, may act as annoyance to the driver [2].
Driving-behaviour-information-based approaches evaluate
the driver’s conduct over a period based on the variations in the
speed, steering wheel angle, acceleration, lateral position and
breaking, there by determining whether the driver is alert or not.
Fig. 1: Aspects that can be used to determine the level of drowsiness of a driver
Liang et al. demonstrated a real-time approach to detect driver [3]
distraction using the eye movements and driving performance,
wherein the data was made to train and test both Support Vector
 2017 International Conference on Intelligent Computing and Control (I2C2)

Table 1: Total vehicle collusion due to fatigue and inattention (2003-2007) [3] III. LITERATURE REVIEW
Total No. of No. of Over the last decade, there have been various studies done
related to drowsiness detection and drunk driving. Features
Year Vehicular collisions due collisions due using a driver’s Visual characteristics, Physiological and
Driving- behaviour based studies have been conducted each
Collisions to fatigue and to fatigue and
having their own advantages for drowsiness detection and by
inattention inattention using sensors for drunk driving detection. Recent Technologies
that have been examined are explained from [1-2, 4, 6-13].
(%) Sinan Kaplan et al., presented a survey that provides a
2003 58,805 8,835 15 comprehensive insight into the well-established techniques for
driver inattention monitoring and introduces the use of most
2004 57,729 9,122 15.5 recent and futuristic solutions exploiting mobile technologies
such as smartphones and wearable devices. The studies were
2005 59,495 8,949 15
categorized into two groups: driver drowsiness and distraction.
2006 58,094 9,837 16.9 A comprehensive compilation, used features, classification
methods, accuracy rates, system parameters, and environmental
2007 55,926 9,797 17.5 details, was represented. A similar approach was also taken for
the methods used for the detection of driver distraction [1].
Ralph Oyini Mbouna et al., presented a visual analysis of Eye
A study conducted by Alcohol & drug Information Centre
State and Head Pose (HP) for continuous monitoring of alertness
(AIDC), India revealed that around 40% of the road accidents
of a vehicle driver. The proposed scheme used visual features
have occurred under the influence of alcohol [5]. A study was such as Eye Index (EI), Pupil Activity (PA), and HP to extract
done at the University of Michigan, clearly shows the effect of critical information on non-alertness of the driver [2]. Mona
Alcohol on Highway Safety which is given in Table 2. Omidyeganeh et al., designed and implemented an automatic
system, using computer vision, which runs on a computationally
Table 2: Effect of Alcohol on Highway Safety [5]
limited embedded smart camera platform to detect yawning.
Implementation of the Viola–Jones algorithm for face and
Blood Alcohol Effect mouth detections and, use of a back-projection theory for
Content (BAC) measuring both the rate and the amount of the changes in the
mouth, to detect yawning along with the histogram of the
Level in g/100ml grayscale image [4]. Anirban Dasgupta et al., proposed a robust
real-time embedded platform to monitor the loss of attention of
0.02 to 0.04 Progressive Deterioration
the driver during day and night driving conditions. The
0.04 to 0.05 Significant involvement in percentage of eye closure was used to indicate the alertness
Accident level. Face detected using HAAR -like features, the eye state
was classified as open or closed using support vector machines
0.02 to 0.04 Dominant factor in accident [6]. Boon-Giin Lee et al., proposed a method to monitor driver
safety for fatigue using two distinct methods: Eye movement
0.02 to 0.04 Accident risk 7-25 times as at monitoring and Bio-signal processing. The monitoring system
BAC 0.01g/100ml was designed on an Android-based smartphone, where it
receives sensory data via wireless sensor network and further
processed the data to indicate the current driving aptitude of the
II. PROJECT OBJECTIVE driver. The sensors used were a video sensor to capture the
driver image and a bio-signal sensor to gather the driver
Development of safety features to prevent drunk and drowsy Photoplethysmograph (PPG) signal. A warning alarm was also
driving, being one of the major challenges for automobile sounded if driver fatigue was believed to reach a defined
industry the objective of this project is to design a system to threshold [7]. G. M. Bhandari et al., presented an efficient
detect and provide driver safety by alerting the driver in case of driver’s drowsiness detection system, by using yawning
drunk condition detection or drowsiness detection. Drowsiness detection. The consideration of eye detection and mouth
detection would be implemented using visual features like Eye detection was done, detecting the driver’s face using YCbCr
and Face detection along with yawn detection all done using method. After that, eyes and mouth positions by using HAAR
OpenCV and Emgu with Visual Studio 2013. For drunk state features. Lastly yawning detection performed by using mouth
detection, an alcohol sensor would be used. In case the driver geometric features [8]. Wang dong et al., aimed at the serious
found to be drunk or drowsy then the alarm would be generated phenomenon of drunk driving in modern society by using a
via a buzzer and using a vibrator to alert the driver. If the alert MCU electronic circuit board in the system along with alcohol
would go off for more than thrice within a specified period, then sensor MQ303A, the alcohol concentration was detected.
Through ADC0809, the detection signal is converted to digital
the engine would be turned off with LED turning ‘ON’ as an
signal, which is handled directly by MCU. As per the digital
emergency indicator.
signal, the car will be controlled automatically wherein it can't
be driven after driver found drunk, thus avoiding the occurrence
 2017 International Conference on Intelligent Computing and Control (I2C2)

of drunk driving [9]. Yue-cheng Wu et al., developed an and the second approach was for Face Detection Using HAAR-
automotive anti-drunk driving system with real-time Like Features [6].
monitoring. The system guarantees the uniqueness of the driver The Algorithm
by combining the function of alcohol detection and face • A web cam is used capture the drivers face in real time.
identification system, putting forward the design of combination • The Face is detected by using HAAR- Cascade
of the alcohol main detection and the image processing auxiliary classifier.
surveillance. It can eradicate the fraudulent conduct that drunk
driving and driver changing [10]. Anjali K U et al., purposed • An ROI is selected on the detected facial image.
this paper to develop a drowsiness detection system. This system • Then Eyes are detected in selected ROI.
works by analysing the eye movement of the driver and alerting
the driver by activating the buzzer when he/she is drowsy. The
system implemented is a nonintrusive real-time monitoring
system for eye detection. During monitoring, the system can
decide whether the eyes were opened or closed. When the eyes
were detected closed for too long, a signal issued to warn the
driver, the system also has an option for making vibration when
drowsiness is detected [11]. T. P. Nguyen et al., demonstrates an
eye tracking system for drowsiness detection of a driver. It is
based on the application of Viola-Jones algorithm and
Percentage of Eyelid Closure (PERCLOS). The system alerts the
driver if the drowsiness index exceeds a pre-specified level [12].

IV. METHODOLOGY
For drowsiness detection, we will be concentrating on the
Visual Feature-based approaches as its proven to provide better
results than physiological and driver behavior methods. Visual
features consist of capturing Eye State, Head Position,
Yawning. Drunk state detection, would be done using an
Alcohol sensor to detect, whether the driver is drunk or not.
A. Eye State
Oyini Mbouna et al., presented a visual analysis of Eye state
and Head Pose (HP) by using a single camera for constantly
Fig. 2: Scheme of Eye and Face detection.
monitoring alertness of the driver. The proposed system would
work in real time finding the eye and pupil centers from a face B. Head Position
object in a live video stream captured by a camera. The detection There are various methods available for object tracking, such
of face and eyes was done using the AdaBoost algorithm and by as mean shift tracking, optical tracking, and Kalman-filter based
adaptive template matching. Then, visual features such as Eye
tracking. The first two methods relied on image intensity values
Index (EI), Pupil Activity (PA), were computed from a video
segment of 4 sec. duration. An SVM classifier was trained with in subsequent frames. As there are illumination dependent, they
the visual features of EI, PA, which was used to learn the driving are not appropriate for our application. As the face size and
patterns of the driver to classify if the subject was either in alert position is known from the detection results of the HAAR
or non-alert state. The non-alert state represented that the driver classifier, a rectangle is circumscribed around the detected face
is either drowsy or distracted [2]. Error deye refers to the relative for subsequent face detection. Using the Kalman-based tracking,
deviation between the estimated and the ground truth pupil the search areas for the face can be reduced; hence, the HAAR
centers, i.e., classifier can detect at a faster rate without loss in accuracy [6].
To track the changes in the head position, Ralph Oyini Mbouna
୫ୟ୶൫หେౢ౛౜౪ ି+ౢ౛౜౪ หหେ౨౟ౝ౞౪ ି+౨౟ౝ౞౪ ห൯ et. al. applied Lucas–Kanade optical flow method. The optical
deye = flow finds an estimate of the feature points between two video
หେౢ౛౜౪ ିେ౨౟ౝ౞౪ ห frames extracted using the good features to track method [2].

Where |Cleft í Ƙleft| and |Cright í Ƙright| denote the C. Yawn Detection
Euclidean distances between the true pupil center positions Yawn detection has a become an integral part in driver
Cleft and Cright and the estimated pupil center positions Ƙleft drowsiness detection. An automatic system, using computer
and Ƙright. The eyes were successfully detected if the relative vision, with a camera to detect yawning has been implemented
error deye is less than a threshold of 0.25 [2]. by Omidyegane et al. [4]. First, detection of the driver’s face
From Anirban Dasgupta et al. proposed system the first would be done using the Viola–Jones algorithm, which has
approach was the estimation of PERCLOS for face detection already been implemented in the OpenCV software library.
 2017 International Conference on Intelligent Computing and Control (I2C2)

Fig. 3: Generalized system diagram of Eye and Face Detection. Fig. 4: Generalized system diagram of mouth detection for yawning analysis.

The AdaBoost algorithm helps to select critical features that play D. Drunk State Detection.
an important role in the classification decision and train the Over the years several attempts have been made to prevent
classifier. On finding the first face on each frame the system drunk driving mainly through a system of laws, enforcement,
stops searching for other faces, thus face search time is reduced. and sanctions. However, these laws are frequently violated.
The next step is mouth detection, where the trained features from Hence by using vehicle-based alcohol detection system,
the XML file are extracted and saved in the camera to reduce the prevents a person with a positive blood alcohol concentration
computational time and make the system more suitable for from starting or operating a vehicle. From Sakairi et al. [13], we
practical operations. The color image for each mouth detected learn that water clusters in an expired gas can be easily
in the video is converted into a grayscale image, then the differentiated into positive and negative charged clusters by
histogram of the grayscale image is acquired. The histogram of applying an electric field. The charged water clusters in the
a normally closed mouth position in the first frame will be saved breath of a person is easily detected by blowing between parallel
as a reference for further calculations. To determine yawning, a plate electrodes, to which a voltage is applied, and a detection
back-projection-theory can also be used. The basic idea in the electrode connected to a pico-ammeter. Small charged water
back-projection theory is to create a similar image giving the clusters are easily deflected by using an electric field to collide
similarity of each pixel of the candidate object to be matched with the detection electrode to generate electric currents, with
(the candidate) with the object of interest (the reference) [4]. this configuration, breath and alcohol peaks for one exhalation
Yawn detection is done to verify the validity of the detected of air can be simultaneously detected.
component. When driver yawns, the mouth starts to open Drunk driving is a major contributor to traffic deaths in India,
resulting in increased threshold pixel value compared to the which is approximately 70 percent of road fatalities. Nearly 10
normal position which is nothing but yawning state, following percent of road accident casualties worldwide are reportedly
are the steps [11]. from India [5]. Wang Dong et al. [9] demonstrates a basic low-
cost system setup using alcohol sensor MQ303A, the sensor
• Select the eye highlighted region. output voltage signal passes through ADC0809 after converting
• Select the mouth highlighted region. the input to MCU. After logic processing of MCU, the driver
• Get Open (yawn) Mouth pixel count (YM) and Normal alcohol level is compared with the calibration value, to control
Mouth pixel count (NM). the car motor to start or stop along with voice warning if the user
• Get Closed (yawn) Eye pixel count (YE) and Normal is above or within alcohol limit. From [10], we learn about a
Eye pixel count (NE). system that can eradicate the fraudulent conduct of finding a
• If (YM >NM) or (YE<NE) replacer to pass the test for starting the car, this is done by using
• Then (if yes: driver is yawning otherwise go back to a CCD camera. On system activation, it captures drivers image,
step 3) image A and stores in MCU then checks driver’s alcohol
concentration in air around driver, if high then the system is
turned ‘OFF’ and sound-light alarm activated, it then
continuously keeps checking the alcohol concentration, once the
level drops to the specified standard then another image
captured, image B.
 2017 International Conference on Intelligent Computing and Control (I2C2)

Both image A and B are verified to check whether they are A. Flow Chart:
identical or not. If identical only then will the engine activate
thus eradicating fraudulent conduct of driver replacement.
V. Principle, Design and Implementation
Developing safety measures to prevent drunk and drowsy
driving is a major challenge for the car industry. Here are some
of the ideas on how to prevent drowsiness and drunk driving
while driving. When a driver enters a car, starts the ignition at
first, he would be told to provide breath sample, by using an
alcohol sensor, it would be detected whether the driver is in
drunk state or not. If yes, then the ignition would turn off. If
driver not drunk, then next step would be face capture. Face
capture would be done by a web camera that would be placed in
front of the driver near dashboard. The camera would capture
the face of driver along with eye and mouth tracking. Eye
tracking is done to detect whether the driver eyes are open or
closed to measure drowsiness level and mouth capture done to
check for yawning while driving. After detection and tracking of
face, eyes and mouth capture while driving, the system
continuously keeps checking for any variations in above
parameters. By using Visual Studio 2013 and OpenCV with
Emgu tracking and detection of facial features is being done.
OpenCV using Emgu being open library contains all XML files
for eye closeness detection and yawn detection would be carried
out using template matching where picture already stored will
be compared to find whether the driver is in a drowsy state or
not. If any one of the parameters gets checked that is if driver
found drunk while driving or in drowsy state then an alarm
would go off and seats would vibrate, thus making the driver
alert again. If the alarm goes off again and again within a certain
time interval, then the system would turn off the ignition and
turn on the indicator lights to warn vehicles coming from
behind to avoid crashing of-vehicles.
Fig. 6: A sequence flow of the events for drowsiness and drunk driving
detection with alert.

Alcohol Detector The flow diagram for ‘Driver Drunk State and Drowsiness
Detection Alert System is as explained in fig.6. The procedure
Engine Start/ Stop of events occurring is explained below;
M Step 1: The driver on entering the car turns ‘ON’ the ignition.
Step 2: To check whether the driver is drunk; this would be done
Web Camera for C Buzzer Alarm manually by the driver by using an alcohol sensor.
Step 3: If driver found drunk that would lead to turning ‘OFF’
Face, Eye, Yawn
the ignition.
Detection U Step 4: If driver not drunk then capture of visual features like
Emergency Lights face, eye and mouth position will begin for drowsiness
(LED) detection. The visual features would be captured in continuous
real-time monitoring, by a web cam.
Step 5: If Eye closure, Head Tilts or Yawn detected this will
lead to activation of buzzer and vibrator, if not then driver
continues to drive.
Fig. 5: A basic system setup for Alcohol Detection. Step 6: If the driver is still found to be drowsy even after
continuous alertness within a short span of time, this would lead
to turning ‘OFF’ the ignition and then the emergency lights
would be turned ‘ON’, to alert other vehicles to avoid collusion.
 2017 International Conference on Intelligent Computing and Control (I2C2)

B. System Diagram References

The basic block diagram that is to be implemented for driver [1] Sinan Kaplan, Mehmet Amac Guvensan, Member, IEEE, Ali
drunk state and drowsiness detection alertness system is as Gokhan Yavuz, and Yasin Karalurt, Driver Behavior Analysis for
Safe Driving: A Survey, IEEE Transactions on Intelligent
shown in fig. 8. Where-in a webcam is used for face and yawn Transportation Systems, Vol. 16, No. 6, December 2015.
detection, an alcohol sensor for drunk state detection. The USB [2] Ralph Oyini Mbouna, Seong G. Kong, Senior Member, IEEE, and
to TTL converter is a device that is used to provide connectivity Myung-Geun Chun, “Visual Analysis of Eye State and Head Pose
between PC and Microcontroller. The microcontroller is used to for Driver Alertness Monitoring”, IEEE Transactions on Intelligent
Transportation Systems, Vol. 14, No. 3, September 2013.
activate the alarm, vibrators, LED’s, and turn ON/OFF the [3] Mario I. Chacon-Murguia and Claudia Prieto-Resendiz, “Detecting
engine when the driver found to be in drunk or drowsy state, Driver Drowsiness: A survey of system designs and technology”,
with the LCD used to display messages regarding the alert. IEEE Consumer Electronics Magazine, October 2015.
[4] Mona Omidyeganeh, Member, IEEE, Shervin Shirmohammadi,
Senior Member, IEEE, Shabnam Abtahi, Aasim Khurshid,
Muhammad Farhan, Jacob Scharcanski, Senior Member, IEEE,
Behnoosh Hariri, Daniel Laroche, and Luc Martel, “Yawning
Detection Using Embedded Smart Cameras”, IEEE Transactions on
Instrumentation and Measurement, Vol. 65, No. 3, March 2016.
[5] T. Sivakumar, Dr. R. Krishnaraj, “Road Traffic Accidents (Rtas)
Due to Drunken Driving in India Challenges In Prevention”,
IRACST- International Journal of Research in Management &
Technology (IJRMT), Vol. 2, No. 4, August 2012.
[6] Anirban Dasgupta, Anjith George, S. L. Happy, and Aurobinda
Routray, Member, IEEE, “A Vision-Based System for Monitoring
the Loss of Attention in Automotive Drivers”, IEEE Transactions
on Intelligent Transportation Systems, Vol. 14, No. 4, December
2013.
[7] Boon-Giin Lee and Wan-Young Chung, Member, IEEE, “Driver
Alertness Monitoring Using Fusion of Facial Features and Bio-
Signals”, IEEE Sensors Journal, Vol. 12, No. 7, July 2012.
[8] G. M. Bhandari, Archana Durge, Aparna Bidwai, Urmila Aware,
“Yawning Analysis for Driver Drowsiness Detection”, IJRET:
International Journal of Research in Engineering and Technology,
Volume: 03 Issue: 02, Feb-2014.
[9] Wang Dong, Cheng Quan Cheng, Li Kai, Fang Bao-hua, “The
automatic control system of anti-drunk-driving”, The National
Natural Science Founds, IEEE, 2011.
[10] Yue-cheng Wu, Yun-qing Xia, Pei Xie and Xiao-wei Ji, The Design
of an Automotive Anti-Drunk Driving System to Guarantee the
Uniqueness of Driver, IEEE 2009.
[11] Anjali K U, Athiramol K Thampi, Athira Vijayaraman, Franiya
Francis M, Jeffy James N, Bindhu K Rajan, “Real-Time
Fig. 7: Basic block diagram for drunk state and drowsiness detection with alert
Nonintrusive Monitoring and Detection of Eye Blinking in view of
system.
Accident Prevention due to Drowsiness”, International Conference
on Circuit, Power and Computing Technologies [ICCPCT], IEEE,
For drowsiness detection, we use a Web Cam interfaced to 2016.
PC to check the visual features of the driver while driving. The [12] T. P. Nguyen, M. T. Chew, S. Demidenko, “Eye Tracking System
detection would be done in Real Time and the web cam output to Detect Driver Drowsiness”, Proceedings of the 6th International
Conference on Automation, Robotics and Applications,
visible on the PC screen. Direct communication will take place Queenstown, New Zealand, IEEE, Feb.17-19,2015.
between PC and microcontroller by using an USB to TTL [13] Minoru Sakairi, “Water-Cluster-Detecting Breath Sensor and App-
converter. If the driver is found to be drowsy, then an alert would lications in Cars for Detecting Drunk or Drowsy Driving”, IEEE
be generated, this would be made by using the buzzer (alarm) Sensors Journal, Vol.12, No.5, May 2012.
and vibrator. For Drunk State Detection, an Alcohol sensor used,
wherein the driver should manually feed input. The relay would
be used as car engine which would start on ignition and stop
working if driver found to be in drunk or drowsy state with the
LED’s used for Emergency lighting to indicate that the car
would be coming to a halt. Based on the alert type, messages
will be displayed on the LCD screen.