International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882

Volume 5, Issue 4, April 2016

An efficient Frame work to identify driver behavior using

Mouth and Eye mapping
Ruchira Padal, Mayuri Pillay, Pradnya Shinde, Dhananjay Gaikwad, Mansi Shinde
Savitribai Phule Pune University, Pune

Abstract
Majority of the road accidents happen due to driver
fatigue which involves symptoms like yawning and
closing of eyes. Thus yawning detection of the driver is
needed. Therefore it is necessary to propose a system
which will thwart road accidents. This system determines
the level of drivers inattentiveness and gives a
forewarning when an approaching hazard appears. The
existing system use face extraction based on support
vector machine (SVM) and circular Hough transform
(CHT) for detecting mouth. The current existing
techniques although are not intrusive, they have several
limitations and are expensive due to the hardware.
Installing a system to alert drivers of their fatigue is a
costly affair. Hence, the proposed system tries to fix this
problem with by using Adaboost Algorithm. The cascade
classifier based on Adaboost algorithm is used for
tracking face components once the face has been
detected.
Keywords: YCbCr color model, Adaboost classification,
chroma components classification, threshold values.

I.

INTRODUCTION

One of the main causes of road accidents is the drivers

fatigue. It can be advocated by the yawning of the driver.
Thus, yawning detection systems are being introduced so
as to detect the drivers tiredness. These systems are
based on the eye and mouth movement of the driver. An
alarm is set off to alert the driver in case yawning is
detected.
In the project, a camera is placed in front of the driver to
capture his/her facial movements. The drivers face is
captured after every predefined time interval and is given
as input to the system. The image captured is segregated
to detect the skin area and then the skin is filtered out.
Light enhancement is done in the segmented image
followed by eye and mouth mapping. The eye and mouth
pixels are counted and threshold values are compared to
set off an alarm, when yawning is detected.
A family of the color spaces that are used in the color
model pipeline in digital photography and video, is the
YCbCr model. It is an encoded RGB information and not
an absolute color space. The Y represents the luminance,

whereas the blue difference and red difference chroma

components are Cb and Cr respectively.
On detection of skin, skin segmentation is done.
Continuous filtering of skin is required for this purpose.
Facial features mapping is done based on this segmented
image. Based on comparison of pixel count the alarm is
set off. Installation of such a system is a costly affair. In
proposed paper, for this purpose, Adaboost algorithm is
used. The AdaBoost algorithm is one kind of selfadaptation iterative algorithm. It selects the most
important features from a big feature candidate set and
makes a weak classifier for every selected one. Then the
multi weak learners are combined to a strong one.
Adaboost is a common way to improvise the correctness
of given learning algorithm. It is generally used to
determine the complication of object detection depending
on cascade structure.
Chrominance also known as Chroma is the signal used in
video systems. It conveys the color information of the
picture, independently from the accompanying luma
signal. Chrominance is represented as two colordifference components: U = B Y (blue luma) and
V = R Y (red luma). Each of these difference
components may have scale factors and offsets applied to
it, as specified by the applicable video standard. In
component YCBCR-encoded video, the chroma is carried
in the Cb and Cr color difference channels of the video
signal.
Threshold
value
is
the
Minimum
or
maximum value conventional for
an feature, characteristic, or constraint) which aids as
a standard for assessment or regulation and any fissure of
which may call for a comprehensive examination of the
condition or the restructure of a system.
Thresholding is a method of translating a grayscale input image
to a bi-level image by using an optimal threshold.
The resolution of thresholding is to excerpt those pixels from
some image which characterize an object (either text or other
line image data like graphs, maps).
Global thresholding defines a solitary threshold for all the
image pixels. When the pixel values of the components
and that of background are equitably unfailing in their
individual values over the complete image, global
thresholding could be used.

www.ijsret.org

181

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 5, Issue 4, April 2016

In adaptive thresholding, dissimilar threshold values for

dissimilar local areas are used.

II.

LITERATURE SURVEY

YCbCr model is more effective in separating luminance

from chrominance, when compared with other color
models like the normalized RGB(rgb) and RGB. Apart
from this, due to the statistical parameters like mean and
standard deviation, that are available in the YCbCr model
to separate out various types of pixels in the image[1].
For segmentation purpose, YCbCr model is used for
elimination of the sensitiveness of illumination[2].
For the extraction of the skin regions of human, YCbCr
color spaced-based skin model has been established,in its
improved version[3].
The AdaBoost algorithm is one kind of self-adaptation
iterative algorithm[4]. It selects the most important
features from a big feature candidate set and makes a
weak classifier for every selected one[5]. Then the multi
weak learners are combined to a strong one. In the
proposed paper, as observed, system installation for
alerting drivers for their drowsiness is a costly affair;
hence, the proposed system tries to fix this problem with
enriched method by using Adaboost. The cascade
classifier based on Adaboost algorithm is used for
tracking face components once the face has been
detected[6].
In this paper chroma subsampling is used for screen
content coding[7]. The paper proposes the methods to
improve the quality of chroma components used.
In this paper chroma vector analysis is used for detecting
particular music part[8]. According to results,mchroma
components of music signals tend to have temporal
correlation in time-frequency representation[9]. Thus
time series of chroma components are analyzed and
attempts are made to segments music parts from the other
parts in the music TV programs.
In this paper a new approach is introduced for Traffic
Sign Recognition. In this, candidate signs are
independently found from two chroma components in the
YCbCr space and then classified into three shape classes:
circle, square and triangle based on measuring the rotated
version correlations.
In the examination of Fingerprint Compression which
uses SPIHT Technique in lieu of Substantial Threshold
Level, to attain acknowledged level of compression, there
is a requirement to compress the fingerprint images at a
noteworthy threshold value.
In this notion, to accomplish great compression ratio,
wavelet based SPIHT algorithm is used for compression
of fingerprint image to get crucial choice of threshold
level.

In the assessment of Real Time Driver Fatigue Detection

Based on Eye and Yawning, the total thresholding
produced on the eyeball color, for developing the eye
pixels onto Y-axis.
In this approach, the dissemination of eyeball pixels on y
axis is presented by disbursing an intensity map and
subsequently, by taking the crags of such intensity map,
we can get the height of the eyeball, which reproduces
how profoundly a person is opening his eyes.
After attainment of the enumerated extent of the opening
level of the eye, distinct threshold to discriminate amid
opening eye and closed eye.
By using threshold value, we get an unbalanced
segmentation of the dark region inside mouth[10].
In the performance Analysis of Image and Audio
Compression Method using Separate Wavelet Transform,
the ECG signal to be flattened is disintegrated to the
anticipated level using the threshold value selection
which should be such that the eminence of the ECG
signal is not prejudiced on restoration and a good amount
of data decrease is also accomplished[11].
As we increase the level of decompositions for a
particular wavelet, we get established numbers of detail
coefficients threshold to zero as there is increase in
threshold value[12].

III.

www.ijsret.org

PROPOSED METHODOLOGY
1. System Overview Diagram

2. The proposed work of our project follows the

below mentioned steps
Skin Detection

182

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 5, Issue 4, April 2016

Using YCBCR model

Skin Area Filtering
Recursion pixel filtering
Edge notation
Face Identification
Face area identification
Face region separation
Face light correction
Eye mapping
Mouth Mapping
Yawning Detection
Mouth Pixel counting
Eye pixel counting
Threshold pixel measurement
Yawning Alert

3.

Algorithm
Input - Image
Output - Face Detection image
Step 0 : Start
Step1 : Read image into Buffered image
Step2 : Skin Extraction by YcBcR method
Step3 : Skin Filtering Recursion pixels filtering
Step4 : Edge Creation by pixel differencing
Step5 : Area Segmentation
Step6 : Area light Enhancement
Step7 : Eye and mouth mapping using Cb and Cr
Chrominance components
Step8 : Face Feature Raster Adjusting
Step9 : Feature presence conformation
Step10 : Mouth pixel count and eye pixel count
Step 11 : Threshold checking
Step 12 : Yawning Alert
Step11 : Stop

4. System Architecture

IV.

RESULTS AND DISCUSSIONS

1. Precision can be defined as the ratio of the

number of relevant yawning alerts raised to the
total number of irrelevant and relevant yawning
alerts identified. It is usually expressed as a
percentage.
A = The number of relevant yawning alerts raised .
C = The number of irrelevant yawning alerts raised .
Precision = ( A/ ( A+ C))*100.
We observe that the tendency of average precision for the
raised yawning alerts are higher compared to other
systems.

Average precision of the proposed approach

2. Recall is the ratio of the number of relevant
yawning alerts are raised to the total number of
relevant yawning alerts identified. It is usually
expressed as a percentage.
A = The number of relevant yawning alerts raised .
B = The number of relevant yawning alerts not raised
.
Recall = ( A/ ( A+ B))*100
We observe that the tendency of average Recall for the
raised yawning alerts are high compared to other system.

Average Recall of the proposed approach

www.ijsret.org

183

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 5, Issue 4, April 2016

V.

CONCLUSION

This paper proposed a new method for detection of driver

drowsiness, by checking the yawing stages. Here in our
proposed model we successfully accomplished the task of
capturing an image from the web cam and store it in
given specific path. The proposed method clearly over
performs SVM method which indicates that the system is
more efficient in yawning detection. The proposed system
designs idea in breakdown model to evaluate yawning
images by using Adaboost algorithm. Moreover, we have
chosen to ignore complicated algorithms in order to
achieve real implementation of the system. On the core
part of the model application successfully detects the eye
and mouth features so that the drowsiness of a person can
be effectively identified.

FUTURE SCOPE
1. The system can be modified for the live cameras
in vehicles.
2. We can enrich the system to get clear pictures
from the camera using the technique called
MIQM (Multi Image Quality Measurement).
3. The system can enrich to detect the yawning even
in night by using Night Vision or Thermal
Cameras.

REFERENCES

Knowledge Discovery, 2008. FSKD '08. Fifth

International Conference, Volume 1, 2008
[7] T. Vermeir; J. Slowack; R. Van Belle; S. Van
Leuven; G. Van Wallendael; J. De Cock; R. Van de
Walle , Guided Chroma Reconstruction for Screen
Content Coding, IEEE Transactions on Circuits and
Systems for Video Technology ,Year: 2015, Volume:
PP, Issue: 99
[8] A. Uemura; J. Katto; K. Higa; M. Ishikawa; T.
Nomura , Music Part Segmentation in Music TV
Programs
Based
on
Chroma
Vector
Analysis,Multimedia
(ISM),
2012
IEEE
International Symposium on ,Year: 2012
[9] T. Huynh-The; H. Nguyen Thanh; H. Tran Cong ,
Traffic Sign Recognition using Multi-class
morphological detection,Advanced Technologies for
Communications
(ATC),
2014
International
Conference on ,Year: 2014
[10]
Narender Kumar, (IJCSIT) International Journal
of Computer Science and Information Technologies,
Vol. 5 (6), 2014, 7821-7826
[11]
Volume 4, Issue 3, March 2014, ISSN: 2277
128X International Journal of Advanced Research in
Computer Science and Software Engineering
[12]
International Journal of Emerging Research in
Management &Technology ISSN: 2278-9359
(Volume-4, Issue-9)

[1] C. E. Premal ;S. S. Vinsley, Circuit, Power Image

Processing based forest fire detection using YCbCr
colour model, Computing Technologies (ICCPCT),
2014 International Conference,2014
[2] Chakraborty ; K. Deb, Bangladeshi road sign
detection based on YCbCr color model and DtBs
vector, 2015 International Conference on Computer
and Information Engineering (ICCIE), 2015.
[3] Xiaopeng Wang; Lucheng Wang ; Tao Lei; Chengyi
Wang, Faced detection based on improved skin
model and local iterated conditional modes, Natural
Computation (ICNC), 2015 11th International
Conference,2015.
[4] Sooyeon Lee; Youngshin Kwak; Youn Jin Kim;
Sehyeok Park and Jaehyun Kim, Contrast-preserved
chroma enhancement technique using YCbCr color
space, Consumer Electronics, IEEE Transactions,
Volume:58, 2012
[5] Jie Song; Xiaoling Lu and Xizhi Wu, An Improved
AdaBoost Algorithm for Unbalanced Classification
Data Fuzzy Systems and Knowledge Discovery,
2009. FSKD '09. Sixth International Conference,
Volume 1, 2009
[6] Kunhui Lin ; Ruohe Yan ; Hong Duan ; Junfeng Yao
and more authors, Objective Classification Using
Advanced Adaboost Algorithm Fuzzy Systems and
www.ijsret.org

184