Shashidhar M.

Sugur, et al International Journal of Computer and Electronics Research [Volume 2, Issue 2, April 2013]

DROWSINESS DETECTION BASED ON EYE SHAPE

MEASUREMENT
Shashidhar M. Sugur
Assistant Professor
IBMR, Chinchwad
Pune(Maharashatra), India
shashi.sugur@gmail.com

Vidyagouri B. Hemadri
Assistant Professor
SDMCET
Dharwad(Karnataka), India
vidya_gouri@yahoo.com

Abstract - Drowsiness of drivers is one of the main

causes of road accidents. Thus countermeasure
systems is required to prevent sleepiness related
accidents. In this paper, a method for detecting
drivers drowsiness and subsequently alerting is
proposed based on eye shape measurement. The
Local Successive Mean Quantization Transform
(SMQT) features and split up snow classifier
algorithm is used detect drivers face and shape
measurement algorithm is used to find the eye blink.
Different criterions such as duration of eyelid
closure, number of groups of continuous blinks and
frequency of eye blink are used to find drowsiness of
the driver. Experimental results show that this new
algorithm achieves a satisfied performance for
drowsiness detection
Keywords - Face Detection, Eye Region Extraction,
Eye Blink Detection, Drowsiness Detection

1. INTRODUCTION
Safe driving is a major concern of societies all
over the world. Thousands of people are killed, or
seriously injured due to drivers falling asleep at
the wheels each year. Recent studies show that
driver drowsiness accounts for up to 20% of
serious or fatal accidents on motorways and
monotonous roads, which impairs the drivers
judgment and their ability of controlling vehicles
[16]. Therefore, it is essential to develop a safety
system for drowsiness-related road accident
prevention.
This paper is organized as follows: in section 2
the related work about the detection of driver
fatigue is presented. Section 3 describes the
proposed method. Experimental results are shown
on section 4 and finally section 5 presents the
conclusion and future studies.

http://ijcer.org

Dr. U.P.Kulkarni
Professor
SDMCET
Dharwad (Karnataka), India
upkulkarni@yahoo.com

2. RELATED WORK
Some research works have been performed in
order to automatically detect drivers fatigue state.
These works for drivers fatigue detection can be
approximately divided into two categories: one
use some physiological characteristics, and
another use some facial image features [3]. The
former detects the change of drivers
physiological characteristics such as brain waves,
heart rate and pulse rate [4-5]. In spite the
detection accuracy is good, but some special
devices must be attached on drivers body. It is
intrusive and causing annoyance to the driver.
People in fatigue show some visual behavior
easily observable from changes in their facial
features like eyes, head and face. The later detects
eyelid movement, head movement, and yawning
based on some image features [6-11]. As eye state
provides significant information, such visual
behavior can be measured and driver's state of
attentiveness can be predicted. Tabrizi presented a
method to estimate the open or close state of eye
by the number of pixels in pupil [7]. Mai Suzuki
proposed to use the open or close degree of eyelid
and blinks time for drowsiness detection [8].
Zhang adopted a vertical projection technique to
locate the position of eyes and judged open or
close state of eye based on gray image feature
[13]. Zhu used Kalman filter to track the eyes and
obtain the parameters of the blink duration which
are used to judge the fatigue of the driver [10].
Danghui Liu finds the difference between
current frames and previous frame [12].
Vidyagouri Hemadri proposed to detect
drowsiness of driver by combining eyelid
movement and yawing [15]. However, all these
algorithms used high definition and high quality
images. In fact, low definition and low quality

ISSN: 2278-5795

Page 108

Shashidhar M. Sugur, et al International Journal of Computer and Electronics Research [Volume 2, Issue 2, April 2013]

images are usually captured by a simple camera.

In this paper we are finding the eye blink based on
shape measurement which can also give good
result for low quality camera.
3. PROPOSED METHOD
In this paper new method to detect the
drowsiness based on eye shape measurements is
proposed. The proposed algorithm consists of four
different phases such as face detection, eye region
extraction and eye blink detection, drowsiness
detection.
3.1 FACE DETECTION

image [1][10]. One of the important properties of

SNoW is the possibility to create lookup-tables for
classification of features.
Consider a patch w of the SMQT feature M(x),
then a classifier
= n weW hxnonface (M(X)) - n weW hxface (M(X))
Equation can be achieved using nonface table
hxnonface and the face table hxface and defining
threshold One single lookup table
hx = hxnonface - hxface

Local SMQT Features and split up snow

classifier detection and face tracking algorithm
[1],[2] is applied for face detection and tracking.
This algorithm is more efficient for tracking than
the Haar Cascade Classifier [14] when working
with multiple image frames, and it can track
different types of facial views, not only the front
view. The size and angle of the face location are
adjusted each time when it shifts.
X is a pixel and D(x) be set of| D(x)|=D pixels
from a local area in an image. D(x) can be vector
or matrix SMQT transformation of the local area
SMQT: D(x)M(x). M(x) is new set of values.
Model of the image can be described as

for single lookup table classification. Let the

training database contain i=1, 2,.N feature
patches with SMQT features. M(x) and the
corresponding classes Ci Non-face table and face
table trained with winnow update rule initially
both table set to 0, Weight set to 1.
There are three training parameters

I(x) = gE(x) R(x) +b

hxface (M(X)) = hxface (M(X)) , for all x w

I(x) =Intensity image

If Ci is a non-face and

R(x)=Reflectance

n weW hxnonface (M(X)) >

E(x)=iiluminance

Then demotion takes place

G=Gain factor and b=bias term

hxface (M(X)) = hxface (M(X)) for all x w

Reflectance feature should be extracted since it

contains object structure. Assume E(x) is partially
smooth. Iiluminance is considered to be constant
in the chosen local area E(x)=E, Vx D SMQT on
the local area will yield illumination and camera
insensitive features. Entire local patterns which
consist of the same structure will result the similar
SMQT feature for certain specified level. Number
of possible patterns using local SMQT features
will be (2l) D.

Procedure is repeated until no changes occur.

Training of non-face performed repeatedly and
finally single table is created.
Feature combinations for one feature, pi,

3.2 SPLIT UP SNOW CLASSIFIER

1 Threshold
2 Promotion parameter
3 Demotion parameter 0< < 1
n we Whxnonface (M(X)) < and Ci is a face then
promotion is conducted as follows

i=1,2,3..(2l) D.
Then,
Vx = (2i)v i=1 | hx(Pi)| for all x w
Let W` (subset symbol) W be a subset chosen to
contain features with largest relevance values.

The SNoW learning architecture is a sparse

network of linear units over a feature space of face
http://ijcer.org

ISSN: 2278-5795

Page 109

Shashidhar M. Sugur, et al International Journal of Computer and Electronics Research [Volume 2, Issue 2, April 2013]

Then,
= n xeW hx (M(X)) can be function as weak
classifier, rejecting no faces within the training
database
The scale and orientation, which are best fit to
the face probability pixels inside the new location,
are selected. As the result, we can know the
square estimation of each face, which is later used
to approximate an axis of the eyes. Figure.1
shows example results of face detection

Figure 3: Binary converted image

Then apply the Sobel edge detection method to
find the edge of the image. Once we find the
binary image dilate the image and remove the
boarder. Then again find the boarder of the dilated
image. Fig4. Shows the dilated image with
boarder removed.

Figure 4: Dilated image

Figure 1: Face detection

3.3 EYE-REGION EXTRACTION
Once we detected human face, next step is to
find the eye region from the human face. As we
know that eyes are present top position of the face.
We simply extract the top half part from the
human face. Where we will find the eyes and
discard the lips and other region of the face. This
will help to speed up the process. Fig2. Shows
result of eye region extraction from face.

Divide image in vertically in to exact two part.

Then take the first half part of the image and find
the maximum axis length and minimum axis
length. If first part is not found then go to second
part of the image and find the maximum axis
length and minimum axis length of the image as
shown in the Fig. 5

Figure 5: Major Axis length and Minor Axis

length
Once we get the major axis length and minor
axis length find the average of the major axis
length and minor axis length.

Figure 2: Eye region extraction

3.4 BLINK DETECTION
We propose an ESD (Eyelids State Detecting)
value, which is a measurement used to classify the
state of eyelid, open or close. The value can be
computed by using the algorithm the objective of
the algorithm is to find the minimum threshold,
which brings the binary image having at least one
black pixel We then threshold the image with the
threshold value (begin with 0). If there in black
pixel found increase the threshold value at least
one white pixel appears. If there is no black pixel
increase threshold value and follow the same
sequence until get black pixel. Fig 3. Show the
binary image of eye region with minimum
threshold value.

http://ijcer.org

Average
Once we get the average from the above
equation set the threshold to check whether eye is
open or closed.
3.5 DROWSINESS DETECTION
Normally, the eye blinks 10 times per minute,
the interval is about two to six seconds, the
duration of each blink is about 0.15 to 0.25
seconds, and the number of groups of continuous
blinks is not more than two times. Therefore, three
criterions are proposed to judge drowsiness.
The first criterion is the duration of eyelid
closure. It is used as distinctive characteristic to
judge whether a driver is drowsy or not. It is
computed by counting the number of eye closed
ISSN: 2278-5795

Page 110

Shashidhar M. Sugur, et al International Journal of Computer and Electronics Research [Volume 2, Issue 2, April 2013]

frames. If the number of related frames is larger

than a threshold drowsiness is reported.
The second criterion is the number of groups of
continuous blinks. It is another characteristic to
judge whether a driver is drowsy or not. The
number of groups of continuous blink can be
computed by the eyelid open and eyelid close.
When driver lies in normal condition, the number
of groups of continuous blinks is not more than
two times, otherwise he may be drowsy.
The third criterion is the frequency of eye blink.
It is also used to judge whether a driver is drowsy
or not. When a driver is drowsy, the frequency of
his blink becomes slower. Numbers of eyelid open
and close are counted continuously to check the
drowsiness of the driver.

5. CONCLUSION
New drowsiness detection algorithm is
introduced in this paper. Local Successive Mean
Quantization Transform (SMQT) features and
split up snow classifier algorithm is implemented
to detect human face. Then eye region is extracted
from the detected face. Eye blink detected using
shape measurement and three different criteria
used to find the drowsiness of the driver.
Experimental results show that this new algorithm
achieves a satisfied performance for drowsiness
detection the algorithm can be extended to work
under various adverse cases such as person
wearing spectacles and under different
illumination conditions.

4. EXPERIMENTAL RESULT
All these algorithms are implemented in
MATLAB 7.11and these programs are tested in
Windows XP 32bit operating system with Intel i3
processors and 3 GB of RAM. This algorithm is
tested with different types of video images.
Videos are converted in to sequence of frames for
the purpose of testing. Table.1 shows the result of
25 videos and Table 2. shows the result obtained
using different criterion.
Table 1: Result of 25 videos
Total Number of Videos
25
Actual Drowsiness in videos
33
Correct Drowsiness Detected
29
False Drowsiness Detected
03
Percentage of correct detection 87%

Table 2: Different criterion

Sl. Criterion used to Number of Correct
No find drowsiness Drowsiness Drowsiness
detected
detected
1
Duration of
13
13
eyelid closure
2
Number of
11
10
groups of
continuous
blinks.
3
frequency of eye
8
6
blink

http://ijcer.org

ACKNOWLEDGMENTS
Authors thank AICTE, New Delhi for financial
Support under Research Promotion Scheme,
8023/RID/RPS-114(Pvt)/2011-12
REFERENCES
[1] Mikael Nilsson, Jorgen Nordberg, and Ingvar
Claesson. Face Detection Using Local Smqt
features And Split Up Snow classifier.
Blekinge Institute of Technology, sweden
[2] Renuka Devi M N, Snehanshu Saha, and
Supriya P Deshpande. Snow Classifier based
Rapid Face Detection and Recognition
Method. In International Conference on
Computer Science and Engineering ISBN:
978-93-81693-81-0
[3] Ann Williamson, Tim Chamberlain. Review
of on- Road driver fatigue monitoring
devices[R]. NSW Injury Risk Management
Research Centre University of New South
Wales. 2005.
[4] Y.Lin, H.Leng, G.Yang, et al. An intelligent
Noninvasive sensor for driver pulse wave
measurement[J]. Sensor Journal IEEE, vol.7,
2007 (5), pp. 790-799.
[5] Chin-Teng Lin, Ruei-Cheng Wu, Sheng-Fu
Liang. EEG-ased drowsiness estimation for
safety driving using independent component
analysis[J]. IEEE Transactions on Circuits
and Systems, vol. 52, 2005 (12), pp.27262738.
[6] M. Saradadevi, Preeti Bajaj. Driver fatigue
detection using mouth and yawning
analysis[C].
International Journal of
Computer Science and Network Security.
Vol. 8, 2008, pp. 183-188.
ISSN: 2278-5795

Page 111

Shashidhar M. Sugur, et al International Journal of Computer and Electronics Research [Volume 2, Issue 2, April 2013]

[7] P. R. Tabrizi, R. A. Zoroofi. Open-closed eye

analysis for drowsiness detection [J]. Image
Processing Theory, Tools and Applications,
2008, pp. 1-7.
[8] Mai Suzuki, Nozomi Yamamoto, Osami
Yamamoto.
Measurement
of
driver's
Consciousness by image processing-A
method for presuming driver's
drowsiness
by eye-blinks coping with individual
differences[C].
IEEE
International
Conference on Systems,
Man and
Cybernetics,vol. 10, 2006, pp. 2891-2896.
[9] Zutao Zhang, Jiashu Zhang. Driver fatigue
etection
based
intelligent
vehicle
control[C].
The
18th
International
Conference on Pattern Recognition, vol.8,
2006, pp. 1262
[10] Zhu Zhenhua, Wu Xiaojuan, Wang Lei, Qi
Lei.Detection method of driver fatigue
basedon blink duration[J]. Computer
Engineering(Chinese), Vol 34, No. 6, 2008,
pp. 201~20.
[11] Phillip Ian Wilson and Dr. John Fernandez
Facial Feature
Detection Using Haar
classifiers JCSC 21, 4 (April 2006)

http://ijcer.org

[12] Danghui Liu, Peng Sun, YanQing Xiao and

Yunxia Yin Drowsines Detection Based on
Eyelid Movement, Second International
Workshop on Education Technology and
Computer Science, 2010[10] Mikael Nilsson,
orgen Nordberg, and Ingvar Claesson Face
Detection Using SMQT Features and Split up
SNOW Classifier, Blekinge Institute of
Technology School of Engineering.
[13] Mohamad Hoseyn Sigari. Driver Hypovigilance Detection based on eyelid
behavior[C].
Seventh
International
conference on Advances in Pattern
Recognition. 2009, pp. 426-429.
[14] Phillip Ian Wilson and Dr. John Fernandez
Facial Feature Detection Using Haar
classifiers JCSC 21, 4 (April 2006)
[15] Vidyagouri B. Hemadri and Umakant P.
Kulkarni. Detection of Drowsiness Using
Fusion of Yawning and Eyelid Movements.
ICAC3 2013, CCIS 361, pp. 583594, 2013.
Springer-Verlag Berlin Heidelberg 2013
[16] Quong Wang, Jingyu Yang, Mingwu Ren,
Yujie Zheng. Driver Fatigue Detection: A
Survey. In Proc Intelligent control and
Automation, Dalion, China, pp 8587- 8591,
2006..

ISSN: 2278-5795

Page 112