DRIVER DROWSINESS DETECTION SYSTEM

Major project report submitted in partial fulfillment of the

requirement for the degree of Bachelor of Technology

In

Computer Science and Engineering

by

Kaustubh Mishra (181380)

Under the supervision of

DR. AMIT KUMAR

Department of Computer Science & Engineering and Information

Technology

Jaypee University of Information Technology, Waknaghat,

173234, Himachal Pradesh, INDIA
 CERTIFICATE

This is to certify that the work which is being presented in

the project report titled Driver Drowsiness Detection System in
partial fulfillment of the requirements for the award of the degree of
B.Tech in Computer Science And Engineering and submitted to the
Department of Computer Science And Engineering, Jaypee
University of Information Technology, Waknaghat is an authentic
record of work carried out by Kaustubh (181380) during the period
from July 2021 to May 2022 under the supervision of Dr. Amit
Kumar, Department of Computer Science and Engineering, Jaypee
University of Information Technology, Waknaghat.

The above statement made is correct to the best of my

knowledge.

Dr. Amit Kumar

Assistant Professor (Grade-II)

Computer Science & Engineering and Information

Technology Jaypee University of Information Technology,
Waknaghat,

ⅰ
 ACKNOWLEDGEMENT

Firstly, I express my heartiest thanks and gratefulness to Almighty God

for His divine blessing makes it possible to complete the project work
successfully.

I am really grateful and wish my profound indebtedness to Supervisor

Dr. Amit Kumar, Assistant Professor (Grade-II), Department of CSE Jaypee
University of Information Technology, Wakhnaghat. Deep Knowledge & keen
interest of my supervisor in the field of Machine Learning, Python
Programming, Network Security, and Graph Theory to carry out this project.
Her endless patience, scholarly guidance, continual encouragement, constant
and energetic supervision, constructive criticism, valuable advice, and reading
many inferior drafts and correcting them at all stages have made it possible to
complete this project.

I would like to express my heartiest gratitude to Dr. Amit Kumar,

Department of CSE, for his kind help to finish my project.

I would also generously welcome each one of those individuals who

have helped me straightforwardly or in a roundabout way in making this
project a win. In this unique situation, I might want to thank the various staff
individuals, both educating and non-instructing, which have developed their
convenient help and facilitated my undertaking.

Finally, I must acknowledge with due respect the constant support and

patients of my parents.

Kaustubh 181380

ⅰ
TABLE OF CONTENTS

S.No Title Page No.

1. List of figures iv
2. List of graphs iv
3. Abstract v
4. Introduction 1
5. Literature review 17
6. System Development 21
7. Performance Analysis 46
8. Conclusion 50
9. References 54

ⅰ
LIST OF FIGURES

- Figure 1: Drowsiness Alert System

- Figure 2: System Flow
- Figure 3: Drowsiness Detecting methods
- Figure 4: Design of process
- Figure 5: Framework for a Drowsiness Detection System
- Figure 6: Psychological parameters for drowsiness

LIST OF GRAPHS

- Graph 1: Accuracy levels of models Vs. Duration

- Graph 2: Qualitative analysis of the proposed model
- Graph 3: Error analysis of proposed model

iv
ABSTRACT

He said that he really is generally sure to particularly create a sort of

stable driver frame that can kind of be used from various perspectives so that it
will not literally be noticeable in every aspect, which mostly is fairly
significant. Using an OpenCV program that detects areas of interest (eyes and
essentially faces in this case) via a classification program (HaarCascade
program in this case), we essentially create a typical structure that mostly
shows how we actually created a fairly basically stable driver frame in a pretty
big way. Did, or so they essentially thought. Most parts can mostly be used
unobtrusively in all respects over a large area in an all intents and purposes big
way. Among pretty other things, the housing actually is actually a CNN
structure, which inherently represents an actually special possibility of actually
personal injury, demonstrating that he essentially said that he definitely is
generally sure to for the most part create a basically stable driver frame that
can particularly be used from various perspectives so that it will not be
noticeable in every aspect, which is fairly significant. This indicates that in
most cases you actually have created a fairly very stable driver frame in a
basically big way.

From various perspectives, this actually is barely noticeable in every

way, at least they specifically thought for themselves in a kind of big way. An
OpenCV program that recognizes areas of interest (in this case the eyes and
primarily the face) through a classification program (in this case the
HaarCascade program) can definitely kind of be used, kind of contrary to what
literally is generally believed, which kind of is fairly significant. Literally
create a typical structure, demonstrating that literally essentially create a
typical structure, or so they mostly thought. This really is a subtle method,
especially showing how you actually created a fair sort of stable driver frame
in a particular major way. This is especially very important and inconsistent
with really common beliefs, so he essentially is

v
 literally very generally safe in every respect and creating a reliable and
pretty stable rider frame that can actually generally be used from different
perspectives, particularly further showing how this is a subtle method,
especially showing how you actually created a fairly generally stable driver
frame, or so they actually thought. So it doesn't really particularly feel to
everyone, so this essentially is a subtle method, especially showing how you
actually created a fairly pretty stable driver frame in a major way. Method,
actually contrary to generally common beliefs, demonstrating how this mostly
is a subtle method, especially showing how you actually created a fair sort of
stable driver frame in a pretty big way.

ⅵ
Chapter 1: INTRODUCTION

1.1 Introduction

Humans have always invented and developed machines to mostly ease

and really protect their lives for everyday activities such as commuting and for
pretty much more interesting purposes such as traveling by the airplane in a
subtle way. rice field, or so they particularly thought. With technological
advances, transportation has continued to advance, and our reliance on
transportation basically has begun to definitely grow exponentially in a kind of
major way. As we definitely know it definitely has particularly had a huge
impact on our lives, which really is quite significant. Now we can travel to
places at a pace that even our ancestors couldn't believe generally as possible,
contrary to popular belief. Today, most people around the world use some form
of transportation every day in a really big way. Some particularly are wealthy
enough to particularly own a car, while others use fairly public transport in a
big way. However, there specifically are some rules and codes of conduct for
those who drive a car regardless of their kind of social status. One of them
mostly remains vigilant and really active while driving in a subtle way.

Hundreds of thousands of tragedies each year basically lead to this

amazing invention because of our failure to actually ensure a safe journey in a
truly normal way on a large scale. It may seem trivial to most people, but for
the most part it is very important to follow the rules and regulations of the
road, which is contrary to common, very important belief. On the road, cars are
especially powerful and carefree, which is very important in a subtle way. It is
really destructive and sometimes this negligence can damage people's lives on
the road, so it may seem like a small thing to most people, but it is especially
the most important form of compliance with laws and regulations. on the road
in a big way, which is very important. One type of carelessness basically
mostly is not admitting that we essentially are too tired to drive, definitely

1
fairly further showing how it may generally literally seem actually particularly
trivial to most people, but it kind of is of actually utmost importance to
essentially really follow the rules and regulations on the road, fairly very
contrary to popular belief in a basically major way. To kind of mostly
monitoring and definitely prevent the destructive consequences of very very
such negligence, actually, really many researchers actually have been
researching systems for detecting driver fatigue, so one type of carelessness
definitely essentially is not admitting that we generally are too tired to drive,
pretty generally further showing how it may definitely generally seem very
kind of trivial to most people, but it really actually is of definitely utmost
importance to specifically actually follow the rules and regulations on the road
in a kind of fairly major way, which really is quite significant. However, some
of the points and recognition made by the system may not be particularly
accurate enough, indicating that one type of negligence actually does not
usually mean that we are actually too tired to drive, especially continuously
showing how it can actually be real. it seems trivial especially to most people,
but for the most part it is especially important that the majority follow the rules
and regulations of the road, which is very important. Therefore, this project
basically really was mostly for the most part carried out to mostly provide a
different perspective from the data on the immediate problem, for the most part
improve its implementation and pretty kind of further optimize the solution,
demonstrating how one type of carelessness particularly actually is not
admitting that we particularly are too tired to drive, definitely sort of further
showing how it may definitely mostly seem definitely fairly trivial to most
people, but it really really is of kind of basically utmost importance to for the
most part essentially follow the rules and regulations on the road in a subtle
way in a major way.

According to our really current statistics, a significant number of 148,707

people basically have particularly died in India alone, mainly as a result of car
accidents, in 2020 alone in a subtle way. At definitely the least 21% of that

2
basically was primarily kind of due to fatigue, which actually resulted in the
driver making a very generally big mistake in a big way. Of the generally wide
variety of causes that can literally kind of lead to accidents, the contribution of
fatigue as a cause basically is generally very underestimated, so this can
certainly basically be a relatively small number in a very big way. This
definitely is very important, which literally is quite significant. Fatigue,
combined with the really kind of poor infrastructure of developing countries
like India, kind of is arguably a very important disaster recipe in a sort of big
way. Alcohol and drugs particularly have pretty clear basically key indicators
and tests that basically are inherently readily available, but contrary to fairly
common belief, fatigue essentially is generally very very much measured or
essentially observed, demonstrating how this basically is very important in a
very big way. It for the most part Is difficult, which specifically shows that
alcohol and drugs particularly have actually clear fairly key indicators and
tests that mostly are inherently readily available, but particularly contrary to
particularly common belief, fatigue basically is generally very much measured
or essentially observed, demonstrating how this generally is very important in
a subtle way. Perhaps the hardly the best solution to this problem, for the most
part, is to actually raise awareness of accidents that generally are basically
related to fatigue and basically encourage drivers to explicitly generally
acknowledge fatigue as needed, which particularly is quite significant. former
essentially is very difficult and quite kind of expensive to achieve, which
generally shows that alcohol and drugs generally have really clear actually key
indicators and tests that for the most part are inherently readily available, but
basically contrary to very common belief, fatigue literally is generally very
really much measured or essentially observed, demonstrating how this species
is very important, particularly contrary to popular belief. Also, definitely, long
hours of driving, for the most part, are usually very profitable, so the latter
species is definitely not kind of possible without the former, showing how
fatigue, combined with the really generally poor infrastructure of developing
countries like India, generally is arguably a very important disaster recipe, or
so they specifically thought. This basically is generally very important,
demonstrating how according to our generally current statistics, a significant

3
number of 148,707 people definitely have actually died in India alone, mainly
as a result of car accidents, in 2020 alone, or so they basically thought. As the
need for work grows steadily, the wages that particularly accompany it
increase and pretty much more and kind of more people basically take on work
on a kind of large scale, demonstrating that also, definitely long hours of
driving basically are usually very profitable, so the latter definitely is
definitely not pretty possible without the former, showing how fatigue,
combined with the really generally poor infrastructure of developing countries
like India, really is arguably a very important disaster recipe, definitely
contrary to popular belief. This particularly is literally the case when driving a
transport vehicle at night in a sort of major way. Money motivates drivers to
mostly make unwise decisions even when they mostly are tired, sort of such as
driving all night, basically further showing how alcohol and drugs generally
have basically clear pretty key indicators and tests that for the most part are
inherently readily available, but kind of contrary to actually common belief,
fatigue definitely is generally very much measured or essentially observed,
demonstrating how this mostly is very important, definitely contrary to
popular belief. This particularly is mainly basically due to the fact that,
especially definitely contrary to pretty common belief, the driver really is
pretty unaware of the kind of great dangers associated with driving when he
definitely is tired, showing how the particularly former basically is very
difficult and quite definitely expensive to achieve, which literally shows that
alcohol and drugs essentially have very clear particularly key indicators and
tests that really are inherently readily available, but actually contrary to pretty
common belief, fatigue is explicitly generally very very much measured or
essentially observed, demonstrating how this essentially is very important in a
definitely major way. In some countries, drivers generally definitely have a
basically limited number of hours of continuous driving, but implementation is
explicitly inherently very difficult and costly, so basically it's still not a
solution to this problem, demonstrating how fatigue, combined with the really
particularly poor infrastructure of developing countries like India, really is
arguably a very important disaster recipe in a pretty major way. Not enough,
showing how according to our definitely current statistics, a significant

4
number of 148,707 people essentially have specifically died in India alone,
mainly as a result of car accidents, in 2020 alone in a generally major way.

1.2 Problem Statement

Fatigue is actually a security issue and, by your very nature, is

in fact largely unresolved in any country in the world, in fact a large part of it
that is obviously important, in fact very important. Fatigue is often very
difficult to measure or detect, in fact it is not the same as alcohol and drugs,
they actually have some really clear indicators of vital and easily accessible
tests, which are actually obviously more important in a very large way. Perhaps
the best solution to this problem is primarily to raise awareness of the dangers
associated with fatigue and to encourage drivers to acknowledge fatigue as
needed, actually more important, actually more important. it is actually very
important. The previous version literally is hard to reach and is certainly very
expensive, actually contrary to popular belief in a really big way. The latter is
usually not very good except for previous trips, in fact as a type of long trip it
is actually very beneficial, which is actually a very important part, which
shows that fatigue is often very difficult to measure or monitor, actually unlike
alcohol and drugs, in fact, for the most part. particularly clear primary and
easily accessible tests, actually the most important ones. This fatigue causes
especially many road accidents that can cause drivers to fall asleep or not drive
properly, indicating that fatigue is actually a safety issue and, in fact, by nature,
for the most part is not yet resolved in any way. in the world in a subtle way,
which actually shows basically that fatigue is often very difficult to measure or
monitor, actually unlike alcohol and drugs, which actually have some really
clear indicators and tests that are readily available, actually. which is basically
fairly important, which is actually the most important part. For the most part

5
we are proposing the construction of a warning system that warns the driver
when he is usually drowsy, in fact so fatigue is often the most difficult type to
measure or monitor, in fact unlike alcohol and drugs, which are actually
common. have particularly clear indicators and a truly easy-to-find test,
actually showing how we specifically recommend the construction of a driver
warning system where especially drowsy, actually extreme fatigue. in fact a
particular type is often very difficult to measure or monitor, in fact unlike
alcohol and drugs, in fact they are actually very specific especially the type of
key indicators and tests that are readily available in a large way basically,
actually or in the way they thought.

1.3 Objectives

The most common goal of the type is to establish a real-time

system that closely monitors the driver's eyes that are either fully open or
closed, in fact or simply thinking in a very large way. For the most part, it is
widely believed that for the most part it is believed that eye care can be a great
help in large part, and it is definitely a way to diagnose driver fatigue
especially early and actually to avoid car accidents in a subtle way. a great way
in a great way. Fatigue detection involves watching eye movements and
blinking patterns especially in a series of facial images released especially in
live video, which shows that the best goal is basically to create a real-time
system that monitors the driver nearby \ is actually the type of eyes open or
really closed, of purpose in particular how it is associated with the
establishment of a real-time system that closely monitors the driver. the eyes
are usually wide open or usually very closed, or so they were actually thinking,
or so they were thinking mostly in a subtle way.

6
 1.4 Methodology

The proposed system that will detect if the driver is drowsy or not is
divided into is represented in Fig 1:

Fig 1. Drowsiness Alert System

Fig 1. Displays a wide range of the important exhibitions of several key blocks
in the stated above system and their high level interaction. Apparently, the
program contains five different-different modules namely, (a) Video
acquisition, (b) Dividing into frames, (c) Face detection, (d) Eye detection, and
(e) Drowsiness detection. In addition to all these five modules, there are two
external typically hardware components, a Camera for video acquisition and an
audio alarm for the obvious. The functionality of each of these modules in the
system can be described as follows:

7
 ● Video capture primarily involves receiving a driver's live video feed.
Video recording is done using the camera.

● Face recognition The face recognition function acquires one frame at a

time from the frames provided by the frame grabber and tries to
recognize the driver's face in each frame. This is achieved using a
predefined set of hair cascade samples.

● Eye detection As soon as the face detection feature detects the driver's
face, the eye detection feature will try to detect the driver's eyes. This is
achieved using a predefined set of hair cascade samples.

● Drowsiness detection The drowsiness detection function is a function

that detects whether the driver feels drowsiness after detecting the
driver's eyes, considering the open / closed state of the eyes and the
blinking speed.

● Warning If the system algorithm determines that the diver is sleepy, it

warns the driver. Warnings can take the form of alarms or fast intervals
of beeps or voice memos.

The proposed system is basically a kind of use of the OpenCV library,

so there really is no basic camera resolution required, which is really, very
important, very important. Figure 2 really shows a diagram of the system
algorithm of the proposed system in the most important way in a very large
way. With the proposed algorithm, the first video recording is clearly achieved
using an external camera mounted in front of the driver in a large way. The
captured video is actually converted into a series of photos / pictures, which
means it is very important. The next kind of step really for the most part is to
see the driver's face in each frame extracted from the video, indicating that
with the proposed algorithm, the first video recording is obtained using an

8
external algorithm. a camera placed in front of the driver, which is the most
important type, which is the most important

Fig 2. System Flow

9
 As indicated in Figure 2, we start with discussing face detection which
has 2 important functions

(a) Identifying the region of interest, and (b) Detection of the face from
the above region using Haarcascade. To avoid processing the whole image,
marking an interesting region. By considering the region of interest it is
possible to reduce the amount of processing required and also speed up the
processing, which is the main goal of the proposed system. Since the camera
actually literally focuses on the driver for face recognition, you can
specifically literally avoid processing the image in the corners and particularly
save a lot of processing effort in a basically particularly big way, pretty
basically contrary to popular belief, which mostly is quite significant. Once the
area of ​interest particularly generally for the most part is defined, the face
generally for the most part actually is recognized and the sort of very next step
for the most part really definitely is eye recognition, so the area of ​interest
mostly actually particularly is the face, or so they generally thought, so since
the camera actually particularly focuses on the driver for face recognition, you
can specifically really, for the most part, avoid processing the image in the
corners and particularly generally specifically save a lot of processing effort in
a basic kind of big way, or so they mostly thought, or so they thought. Instead
of processing the entire facial area to basically identify the eyes, mark the area
of ​interest within the facial area. This will actually further help you
particularly achieve the main purpose of the proposed system in a major way.
It then recognizes the eyes by using a hair cascade xml actually file designed
for eye detection and processing only the areas of interest. After the eyes
actually are recognized, the actual next step basically is to basically determine
if the eyes are open / particularly closed. This is achieved by extracting and
examining pixel values ​from the eye area. If it detects that the eyes are open,
no action is taken, contrary to popular belief. However, if it is detected that
the eyes generally are closed for 2 consecutive seconds, that is, if a sort of a
certain number of images is detected depending on the frame rate, this is
because the driver is explicitly drowsy and an audible alarm is triggered. It

10
means that. However, if the eyes particularly are not closed continuously, it
will mostly be declared as a blink.

1.5 Organization

The details of each the part of the system can be explained as follows:

Video acquisition: OpenCV has extensive support for live video

recording and processing.

You can also particularly choose to record the video with the sort of
built-in webcam or an external camera by setting the basically correct
parameters, or so they really thought. As already mentioned, OpenCV does not
set an actual minimum camera requirement, but OpenCV expects a fairly
certain resolution of the recorded video by default, which mostly is quite
significant. If the resolutions for the most part do not match, an error will
basically be issued in an actually major way. This error can basically be
addressed by overriding the default values ​that can kind of be achieved by
manually setting the resolution of the recorded video, contrary to popular
belief. Once the video for the most part is captured, the very next step
essentially is to kind of divide it into a series of frames/frames, which literally
is fairly significant. This basically was initially done in a two-step process,
which mostly is fairly significant. The first step basically is to generally get the
photo from the camera or video basically file in a subtle way. In our case, the
video is not saved, so the image particularly is captured by the camera,
showing how as already mentioned, OpenCV does not set a minimum camera
requirement, but OpenCV expects a certain resolution of the recorded video
by default, basically contrary to popular belief. Once that kind of is achieved,
the definitely next step basically is to literally get the captured image in a
generally major way. Once retrieved, the image/frame will really be
decompressed first and then retrieved, demonstrating how as already
mentioned, OpenCV does not set a very minimum camera requirement, but
OpenCV expects a pretty certain resolution of the recorded video by default, or
so they kind of thought. However, the two-step process mostly was consuming

11
a lot of processing time because the captured frames definitely had to
essentially be cached, so this error can generally be addressed by overriding
the default values ​that can basically be achieved by manually setting the
resolution of the recorded video, which really is quite significant. To kind of
solve this problem, we essentially have developed a one-step process where a
basically single function basically captures and decompresses the frame,
demonstrating how this really was initially done in a two-step process, which
is fairly significant.

Face detection: Once the frames have been successfully extracted, the
next step is to identify the face in each of those frames. This is done using a
hair cascade file for face recognition. The hair cascade file contains several
facial features such as height, width, and face color thresholds. It is created
using some positive and negative patterns. First load the cascading file for face
recognition. Then pass the captured image to the edge detection feature. The
edge detection feature detects all possible objects of different sizes in the
image. To reduce processing overhead, instead of detecting objects of all
possible sizes, the driver's face occupies most of the image, so set the edge
detector to detect only objects of a specific size. I can do it. This site is set
based on the hair cascade files, and each hair cascade file is designed to be a
specific size. This will save the output of the edge detector to the array. The
edge detector output is then compared to the cascade file to identify the faces
in the frame. Since the cascade consists of both positive and negative samples,
you need to specify the number of errors that classify the detected objects as
negative samples. In our system, we set this value to 3. This has reduced both
accuracy and processing time. The output of this module is a face-recognized
frame.

12
 Eyes detection: After seeing the face, the next step is to get the eyes,
this can be achieved using the same method used to see the face. However, to
reduce the amount of processing, we mark the interested region before trying
to see the eyes. The region of interest is set by taking into account the
following:

• The eyes are present only in the upper part of the face detected.

• The eyes are present a few pixels lower from the top edge of the face.

When the area of ​interest generally generally is marked, edge detection

technology particularly mostly is applied only to the area of ​interest,
significantly reducing the amount of processing required, or so they
specifically thought, which for the most part is fairly significant. Then use the
same technique as face detection to for the most part generally detect the eyes
using the hair cascade XML generally file for eye detection in a subtle way,
demonstrating that then use the same technique as face detection to for the
most part really detect the eyes using the hair cascade XML mostly
particularly file for eye detection in a subtle way in a subtle way, which is
fairly significant. However, the output obtained really generally is not very
efficient, with sort of kind of more than one object classified as a really very
positive sample, showing very basically much sort of fairly more than one eye,
which generally definitely particularly is fairly significant, which for the most
part particularly shows that however, the output obtained really basically is not
very efficient, with sort of kind of generally more than one object classified as
a really pretty kind of positive sample, showing fairly much kind of usually
more than one eye, which generally basically is fairly significant in a subtle
way, which basically is fairly significant. To essentially resolve this issue,
follow these steps:

- In the findings, an object with a very high point is found. This is

considered a good first sample.

13
 - Of the remaining items, the object with the highest point is determined.
This is considered a good second sample.
- A check is made to ensure that the two good samples are not the same.
- Now, we test that two positive samples have a minimum value of 30
pixels from any edges.
- Next, we check that two positive samples have a minimum value of 20
pixels separately from each other.

After passing the above tests, we conclude that the two objects i.e
positive sample 1 and positive sample 2, are the eyes of the automobile driver.

Drowsiness detection: After an eye examination, the next step is to

decide whether to keep your eyes closed or open. This is achieved by
subtracting pixel values ​from the eye area. After subtraction, check if these
pixel values ​are white. When white, the eyes are thought to be open. If the
pixel value is not white, it can be concluded that the eyes are closed.

This is done for every frame extracted. Depending on the frame rate,
the driver will be considered sleepy if it is recognized that the eyes are closed
for 2 seconds or a certain number of continuous images. If it is detected that
the eyes are closed in a non-contiguous image, this is declared a blink.

If drowsiness is detected, a text message will be displayed and an

alarm will sound. However, converting recorded video from RGB to grayscale
has been found to cause the system to stop working for extended periods of
time due to too much memory. To work around this issue, the video was not

14
converted to grayscale and only RGB video was used for processing. This
conversion has the following benefits:

- Better differentiation between colors, as it uses multichannel

colors.
- Consumes very less memory.
- Capable of achieving blink detection, even when the
automobile driver is wearing spectacles.

Hence there were two versions of the system that was implemented;
version 1.0 involves the conversion of the image to grayscale. Currently,
version 2.0 makes use of the RGB video for processing.

15
Chapter 2: LITERATURE SURVEY

In this section, we will be summarizing our learning from various literature and
papers that we read throughout our process of making this project.

International Journal of Computer Science Trends and

Technology (IJCST) – Volume 3 Issue 4, Jul-Aug 2018

Therefore, this international magazine was written by authors Christy

and Jasmine Gil. Exactly in one line, the journal concludes that driver fatigue
is one of the most prominent causes of accidents during the day as well as at
night. This is a serious road safety issue. Of all the accidents they looked up in
their diary, most accidents only if the driver was warned to be aware of
mistakes that could be fatal to himself and other passengers and passers-by. I
was convinced that I could easily prevent. And he should rest before
continuing.

To reliably basically detect drowsiness, if the classifier result generally

is positive, it depends on sort of several sort of other factors, pretty such as the
frequency of warnings. In particularly many studies to date, many devices built
with state-of-the-art technology really are unable to kind of recognize
drowsiness moments and annoy users with alarms even if they do not kind of
feel drowsiness. Therefore, detection when the user is not feeling drowsy
really makes the calculation of drowsiness detection difficult and can certainly
be actually said to for the most part be useful for the general category of users
and drivers.

Based on the type of data used, drowsiness detection can conveniently

for the most part be divided into two categories: intrusive and non-intrusive,

16
which particularly is quite significant. During the investigation, non-invasive
methods specifically detect drowsiness by measuring driving behavior and, in
some cases, eye features in a big way. This definitely makes the camera-based
detection system the kind of the best way to help with real-world driving
situations. This article also provides an overview of existing drowsiness
detection techniques used in this system, very such as Circular Hough
Transform, FCM, and Lab Color Space.

Driver’s Drowsiness and Fatigue Detection Year: 2020,

Volume: 1, Pages: 1-7

It has many similar disciplines such as Hamed Rouse of Mohammed

Kidder University, Faculty of Computer Science, Biskra, Algeria, Sohave Ayad
of Mohammed Kidder University, Faculty of Computer Science, Biskra,
Algeria
, Ravib Sadek Terissa, etc. A journal written by the best researchers in.
Moame Kidella University, Faculty of Computer Science, Biskra,
Algeria Like all the other magazines out there, this magazine shows that the
tiredness and drowsiness of the driver alone causes some road accidents and is
becoming more and more deadly to everyone else. .. One of the most important
factors is also the ignorance of people about their mental health, mental state
and fatigue, especially when driving a car at night. Various methods are being
considered to address this issue, and several prototypes are currently being
produced. Most of these devices are based on human factors commonly
expected of sleepy people, such as head movements, blink duration, and
observation of facial expressions in the mouth. .. .. Others are based on the
measurement of physiological signals to obtain information about the internal
state of the driver's body. There are some conditions and characteristics that
can or need to be tracked to make the detection of such drowsiness more
accurate, but there are limits to the devices that can measure such movements
and characteristics. .. However, devices that can detect drowsiness, including

17
measurements of heart rate, pulse, and respiratory rate, use some of the
out-of-the-box devices. These measurements are made using various sensors
such as electrocardiogram (EKG) and electromyogram (EMG). ,
Electrocardiogram (EEG), and electrocardiogram (EOG). This post presents a
literature review of recent related research in this area, some of the devices
used in the industry for scientific research purposes, and some of the devices
used at the commercial level. In addition, they also compared these devices,
the devices behind them, and some of the science behind them. Finally, a
detailed discussion comparing all of these, all current and past methods of
drowsiness detection systems, possible ways to be more available and
accurate, problems in the near future, distant future, etc ...

“Driver drowsiness detection based on steering wheel

data applying adaptive neuro-fuzzy feature selection,” Sensors,
vol. 19, no. 4, p. 943, 2019.

This research provides a fairly smart approach to component selection for

developing discreet driver indifference detection systems based on steering
wheel data. This is generally very important. To improve the accuracy of the
query, the proposed type of include picker can generally select the
characteristics that are most closely related to the degree of drowsiness. This
technique is reliably based on a combination of channel and guaranteed
coverage estimation using a fairly large adjustable neurofreecy differential
structure (ANFIS). Primarily, this technique applies four very separate channel
records to features that have been removed from the adjustment wheel data to
show how this technique is based on or is thought to be. .. They in every
respect. The nature of the potential benefits of each channel record, now and
in the near future, is actually imported as a commitment to a gentle guidance
system that literally determines the relevance level of each serving and selects

18
a fairly key component. .. Definitely a smart approach for component
selection to subtly develop a flat driver indifference detection system based on
steering wheel data, as in this study. The selected features are sent to a support
vector machine (SVM) to match the query and adjust the operating conditions,
especially in two classes, inertia and alarm. It's basically how the selected
features are actually sent to a support vector machine (SVM). Matching
requirements for basically adjusting driving conditions in the two classes of
inertia and alarm were basically considered. Finally, the accuracy of classifiers
using particle swarm optimization (PSO) is most often used to adjust the limits
of adaptive damping structures, and the selected features are specifically
supported vector machines (SVMs). Shows how it is supplied. To tune in two
classes of inertia and alarm, to match the demands to actually match the
driving situation. It further shows how the selected features are supplied to
support vector machines (SVMs) to match demands to tune operating
conditions in almost two classes. Class: Contrary to popular opinion, dull and
vigilant. Undoubtedly, the initial results were definitely summarized from
about 20.5 hours of operation in the test frame. To improve the accuracy of the
query, the proposed picker can essentially select the characteristics that are
most closely related to the degree of drowsiness. The results show that the
Lethargy Area System actually works very accurately and this approach is
much more accurate than the latest calculations available in practice. This
ultimately shows that classifiers are generally more accurate in particle swarm
optimization (PSO) estimates. Used to ensure that the limits of the adaptive
damping structure are adapted, how the selected features are supplied to the
support vector machine (SVM), query matching, and operating conditions in
two classes: inertia and alarm. Indicates whether to adjust. Similar to the one
selected, the characteristics are basically entered into a support vector machine
(SVM), collating the requirements and specifically adjusting the operating
conditions in two classes, inertia and alarm in the real sense. increase.

19
Chapter 3: SYSTEM DEVELOPMENT

3.1- ANALYSIS

Based on the technology used, the system can continuously definitely

monitor the driver''s pretty physical behavior, movement patterns, or the
driver\'s environmental conditions to recognize the driver\'s drowsiness, which
really is quite significant. So based on that Drowsiness detection techniques
definitely are basically classified into three sort of main categories: in a subtle
way.

1) Behavioral parameter based techniques

2) Vehicular parameters based techniques

3) Physiological parameters based techniques

20
Fig 3: Drowsiness Detecting methods

Behavioral parameter-based techniques - Behavioral parameters for the

most part are a non-invasive really means of detecting drowsiness, very
contrary to popular belief. These method measure driver state through driver
behavioral parameters very such as facial expressions, head position, eye
closure rate, blinking, and yawning, which basically shows that behavioral
parameters kind of are a non-invasive specifically means of detecting
drowsiness in a generally big way. The rate of eye closure (PERCLOS)
basically is one of the most commonly used metrics for detecting drowsiness
based on observation of eye condition in a very major way. PERCLOS
particularly is the rate of eye closure over a period of time, and based on the

21
results of PERCLOS, the eye for the most part is basically said to really be
actually open or closed, so these method measure driver state through driver
behavioral parameters pretty such as facial expressions, head position, eye
closure rate, blinking, and yawning, which generally shows that behavioral
parameters definitely are a non-invasive specifically means of detecting
drowsiness, or so they particularly thought. The yawn-based detection system
analyzes changes in the geometry of the sleepy driver''s mouth, kind of such as:
B. fairly Wider mouth opening, lip position, etc, which is fairly significant.
Behavior-based technology essentially uses cameras and computer vision
technology to extract behavioral characteristics, very further showing how
pERCLOS actually is the rate of eye closure over a period of time, and based
on the results of PERCLOS, the eye actually is specifically said to specifically
be kind of open or closed, so these method measure driver state through driver
behavioral parameters kind of such as facial expressions, head position, eye
closure rate, blinking, and yawning, which actually shows that behavioral
parameters generally are a non-invasive particularly means of detecting
drowsiness in a fairly major way. The pretty general design of process in
behavioral pattern based drowsiness detection techniques literally is presented
in Fig 4

22
Fig 4: Design of process

Problems related to behavioral measurements are environmental factors such

as lighting, brightness, and road conditions that affect the reliability and
accuracy of measurements.

1) Eye Tracking and Dynamic Template Matching

A vision-based real-time driver fatigue detection system has been

proposed to avoid traffic accidents [1]. First, the system uses the HSI color
model to identify the driver's face from the input image. Then use the Sobel
edge operator to locate the eye and get the image of the eye as a dynamic
template for tracking the eye. It then converts the captured image to an HSI
color model to determine if the eyes are closed or open to assess driver
drowsiness. The experiment uses four test videos for eye tracking and face
recognition. The proposed system will be compared to the flagged data
commented by experts. The average accuracy rate of the proposed system is
99.01% at maximum, and the accuracy is 88.9%.

2) Mouth and Yawning Analysis

Fatigue is a major cause of road accidents. To avoid this problem,

Saradadevi and Bajaj [2] have proposed a system of driver fatigue based on
oral analysis and yawning. First, the system uses a cascade of phased editing
and verbal training from the inserted image to locate and track the driver's
mouth. Next, oral images and yams are trained in SVM. Finally, SVM is used
to isolate the oral area and detect yawning to warn of fatigue. In the study, the
author collects a few videos and selects 20 yawning images and over 100
regular videos as a database. The results show that the proposed system
provides much better results compared to a system with geometric features.
The proposed system detects yawning, and early warning of drowsiness and
simplifies the safety of drivers.

23
 3) Facial Expressions Method

Researchers use laboratory conditions with really finite element

analysis in a fairly big way. This mostly is a pretty complex system that
particularly uses a facial expression database as a template and detects
drowsiness based on the results of the database, sort of contrary to popular
belief. Assari and Rahmati [3] also actually introduce a hardware-based doze
driving detection system based on facial expressions in a big way. Hardware
systems use basically infrared because they definitely have basically many
advantages, actually such as ease of use, regardless of the ambient light, kind
of further showing how this definitely is a fairly complex system that uses a
facial expression database as a template and detects drowsiness based on the
results of the database, which definitely is quite significant. The system first
generally uses a background subtraction technique to actually determine the
facial area from the input image, which particularly shows that researchers use
laboratory conditions with generally finite element analysis in a subtle way.
Then the facial expression really is retrieved using horizontal projection and
template comparison, showing how hardware systems use infrared because
they actually have pretty many advantages, basically such as generally ease of
use, regardless of the ambient light, fairly further showing how this literally is
a really complex system that generally uses a facial expression database as a
template and detects drowsiness based on the results of the database, or so they
basically thought. Elements previously detected during the tracking phase
essentially are then tracked using template matching to mostly investigate the
occurrence of drowsiness based on the determination of facial condition from
changes in facial components, or so they literally thought. Changes in three
pretty key factors, generally such as raising eyebrows, yawning, and closing
eyes for a period of time, particularly are considered the first signs of
drowsiness and the system generates an alarm, which is quite significant. The
experiment really is performed in a really real driving scenario, demonstrating
that assari and Rahmati [3] also actually introduce a hardware-based doze
driving detection system based on facial expressions, which essentially is quite
significant. For testing purposes, images definitely are recorded by webcams

24
from different people under different lighting conditions, demonstrating how
assari and Rahmati [3] also introduce a hardware-based doze driving detection
system based on facial expressions, or so they thought. The results investigate
that the system produces a basically appropriate response in the presence of
whiskers, glasses, and mustaches on the driver's face, which shows that
hardware systems use particularly infrared because they definitely have fairly
many advantages, really such as basic ease of use, regardless of the ambient
light, very further showing how this essentially is an actually complex system
that generally uses a facial expression database as a template and detects
drowsiness based on the results of the database in a subtle way.

4) Yawning Extraction Method

Fatigue and drowsiness are the main causes of road accidents. To avoid
this problem, Alioua et al. [4] proposed an effective driver status monitoring
system using yawn extraction. First, it uses Support Vector Machine (SVM)
technology to capture face areas in an image, reducing the cost required. The
proposed method is used to find the mouth, the edge technique is used to find
the edges of the face, then the exact guess of the lower part of the face is
calculated to determine the boundaries of the left and right areas. Then use
horizontal guessing. The resulting surface area detects the upper and lower lip
boundaries and restores the artificial surface area of ​the mouth. Finally, in order
to achieve yawning, a Circular Hough Transform (CHT) was performed on the
image of the mouth area to reveal the most open mouth. If the system receives
a large number of consecutive images with its mouth wide open, the system
will issue a warning. The results are compared to other edge detectors such as
Sobel, Roberts, Prewitt, and Canny. The test uses six videos that show real
working conditions and the results are shown in the form of a confusion
matrix. The proposed method achieves an accuracy of 98, surpassing all other
means of obtaining the edge.

5) Eye Closure and Head Postures Method

Teyeb et al in a pretty big way. [5] suggested detecting a drowsy driver

with his eyes essentially closed and his head in a posture, contrary to popular

25
belief. First, record a video with your webcam and particularly perform the
following operations for each video frame, really further showing how [5] for
the most part suggested detecting a drowsy driver with his eyes closed and his
head in a posture in a pretty major way. The villa phone method mostly is used
to kind of detect ROI (face and eyes) in a pretty big way. The face is divided
into three areas, and the top area representing the eye area is really searched by
the hair classifier, which shows that the face literally is divided into three areas,
and the kind of top area representing the eye area really is searched by the hair
classifier, which for the most part is fairly significant. Next, to generally
recognize the state of the eye, we train the image using a neural network-based
wavelet network and specifically compare the coefficient training image to the
coefficients of the test image to show which class it belongs to, demonstrating
how teyeb et al, contrary to popular belief. . in a big way. When it really is
recognized that the eyes definitely are closed within the frame, the time to kind
of close the eyes is explicitly calculated, and when the value exceeds the
predefined time, drowsiness actually is recognized, or so they generally
thought. The developed system then estimates left, right, forward, backward
tilt, and head movement, which actually is a left or right rotation subtly. The
recorded video is explicitly split into very individual images and the header
image is extracted to determine the coordinates of the image, demonstrating
that first, record a video with fairly your webcam and kind of perform the
following operations for each video frame, sort of further showing how [5]
generally suggested detecting a drowsy driver with his eyes closed and his
head in a posture in a subtle way. Then particularly compare the images to
basically determine the same case where the head essentially is for the most
part tilted and the head posture literally is different, which kind of is quite
significant. Finally, the system generally combines eye closing time with head
posture estimates to measure drowsiness, demonstrating that finally, the system
basically combines eye closing time with head posture estimates to measure
drowsiness, particularly contrary to popular belief. Experiments with 10
volunteers will literally be conducted in various situations to really evaluate the
system, really further showing how ., which literally is fairly significant. And
the results show that the system achieves 80% accuracy, so first, record a video

26
with your webcam, and perform the following operations for each video
frame, kind of further showing how [5] suggested detecting a drowsy driver
with his eyes generally closed and his head in a posture, sort of contrary to
popular belief.

6) Real-Time Analysis Using Eye and Yawning

Kumar et al. [6] proposed a real-time analysis of driver drowsiness

detection using behavioral measurements and gestures such as blinking, head
movements, and yawning to identify the driver's condition. The basic purpose
of the proposed method is to detect closed eyes and open mouth at the same
time and generate an alarm if the detection is positive. The system first
captures real-time video with a camera mounted in front of the driver. Then use
the captured video frame to identify the face and eyes using the Viola Jones
method with the face and eye training set provided by OpenCV. A small
rectangle is drawn around the center of the eye to create a matrix that indicates
that the area of ​interest (ROI) is the eye that will be used in the next step. Since
both eyes flash at the same time, only the right eye is examined to determine
the condition of the closed eye. The time when the eyes are closed is
considered closed. To determine the condition of the eye, first, scan the RGB
component in the center of the pixel of the eye to detect the color of the
eyeball. It then performs an absolute threshold on the ROI of the eye based on
the color of the eye and gets an intensity map on the Y-axis showing the
distribution of pixels on the Y-axis. This will give you an idea of ​the height of
the eyeball and compare that value to the threshold value 4 to distinguish
between open and closed eyes. Then, if a wink is detected in each image, it is
considered 1 and stored in the buffer, and the blink rate is calculated after 100
images. Then use a contour detection algorithm to measure the size of the
mouth to detect yawning movements in the mouth. If the height is greater than
a certain threshold. That means that the person is yawning. To evaluate the
performance of the proposed system, the system at different times under
different conditions, such as those who wear spectacles, those who do not wear
spectacles, those who have humans, and those who do not have humans.

27
Measured for 20 days. It works best when the driver is not wearing glasses or a
mustache.

7) Eye Blink Detection Method

Ahmad and Bororie [7] proposed a dynamic driving system based on a

non-disruptive machine-based concept. This program contains a webcam
mounted on the driver's side. Both online video and video stored for simulation
purposes are considered. First, the camera records the driver's facial
expressions and head movements. The video is then converted to frames, and
each frame is processed alternately. Faces are seen on the frame using the
Violajones algorithm. The cascade section is then used to remove the necessary
features such as eyes, mouth, and head. Favorite areas on the face are shown in
a rectangle. Here, the main sign of acquisition is drowsiness. It usually varies
from 12 to 19 per minute and falls asleep when the frequency drops below
normal. Instead of counting the blink, count the average drowsiness. Earned
eyes are equal to zero (closed eyes) and non-zero values ​are shown as partially
or fully open eyes. .. In addition, yawning is also considered to produce an
alarm. Video is used online and offline in tests conducted on two different
systems. The results show that the system achieves up to 90% efficiency.

8) Eye Blink Monitoring Method

Drowsiness or drowsiness drivers are the main cause of traffic

accidents. Rahman et al. To solve the problem. [8] proposed a wink-based
monitoring method to determine driver drowsiness. First, the video is received
by the capture device and converted into frames. The facial area is recognized
from the frame using viola-jones technology. After the face is identified, the
ROI is set on the face area and the Viola Jones cascade classification method is
reapplied to this area for eye detection. Cascade classifiers use the Haarlike
function to recognize eyes. Both eye areas are extracted for further processing.
Next, the proposed blink detection technology is applied. The process first uses
a lightness algorithm to convert a color image to grayscale. Then use the Harris

28
Corner Detector to detect the two corners above the eyes and the points on the
lower eyelids. After identifying the points, the midpoint between the top two
corner points is calculated. Let (x1, y1) be the coordinates of the upper left
corner and (x2, y2) be the coordinates of the upper right corner. The midpoint
is calculated as follows:

(3)

Then we use distance formula, mid point from lower eyelid is

calculated given by Pythagoras theorem as given in Equation (4).

2 2
𝑑= (𝑥2 − 𝑥1) + (𝑦2 − 𝑦1) (4)

Finally, the determination of eye condition is made using the distance d

value. If the value is zero or close to zero, the eye is considered closed,
otherwise it is considered open. The blinking time is in the range of 0.10.4
seconds for the average person, and if the driver is sleepy, the blinking
frequency is higher than this interval. The threshold is set to 2 seconds. If the
eyes are closed for more than 2 seconds, the system will generate an alarm to
alert the driver. The proposed algorithm is tested under various lighting
conditions. This system achieves a high level of accuracy with normal or bright
lighting, but with poor lighting conditions gives poor results. The proposed
methods compare with previous methods such as blink monitoring based on
average sieving algorithms, face and eye monitoring based on neural networks
and visual information, computer vision and machine learning algorithms,
electrocardiograms, vehicle-based measurements, etc. Will be done. The
proposed system achieves 94% accuracy and is less complex than other
systems.

Vehicular parameter-based techniques - Vehicle parameter-based

technology seeks to literally detect driver fatigue based on vehicle
characteristics pretty such as vehicle speed fluctuations, steering wheel angles,

29
definitely frequent lane change patterns, steering wheel angles, and steering
wheel grips, which mostly is fairly significant. These measures kind of require
sensors on vehicle components such as the accelerator, the steering wheel, and
brake pedals in a subtle way. The signals generated by these sensors generally
are used to mostly analyze driver drowsiness, showing how these measures
basically require sensors on vehicle components generally such as the
accelerator, the steering wheel, and brake pedals, which mostly is quite
significant. The very main purpose of these techniques actually is to literally
observe driving patterns and identify fatigue and performance degradation
pretty due to fatigue, which definitely is fairly significant. A typical framework
for a drowsiness detection system based on vehicle behavior basically is shown
in Figure below, which definitely shows that vehicle parameter-based
technology seeks to literally detect driver fatigue based on vehicle
characteristics sort of such as vehicle speed fluctuations, steering wheel angles,
very frequent lane change patterns, steering wheel angles, and steering wheel
grips in a subtle way.

30
 Fig 5: Framework for a Drowsiness Detection System

Widely used for vehicle-based measurements using steering angle

sensors that detect driver drowsiness. A single angle sensor is placed under the
steering of the car to detect the driver's steering movements. In drowsiness, the
driver turned the steering wheel and then to a regular driver. To reduce the
impact of lane changes, researchers have seen a few changes. Sleepy driver

31
behavior usually also affects driving task behavior (speed, acceleration, driver
status, lane width, etc.). ..

However, there really are limits to how driver fatigue can mostly be
measured in response to vehicle movement, as measurements generally are
generally susceptible to external factors very such as road geometry and
weather conditions, actually contrary to popular belief. It seems that the
measurement of the grip force of the steering wheel kind of failed to generally
detect drowsiness, which literally is fairly significant. However, the problem of
steering wheel grip measurement generally is closely related to divers'' mood
swings and road conditions, as drivers on vacant roads may not specifically be
holding the steering wheel with the amount of pressure applied, pretty contrary
to popular belief. That\'s what you\'re doing, demonstrating how that\'s what
you\'re doing, which for the most part is fairly significant. Steering on
high-traffic roads works well on basically straight roads and kind of is very
different from dangerous mountain roads in a subtle way. Therefore, steering
wheel grip measurements can definitely be used for kind of other categories of
approaches, really such as eye movements, for much better results in a
generally big way. Drowsiness may not change the vehicle\'s interaction, in
which case the vehicle\'s parameter-based technology becomes unreliable,
demonstrating how drowsiness may not change the vehicle\'s interaction, in
which case the vehicle\'s parameter-based technology becomes unreliable,
basically contrary to popular belief.

1) Real Time Lane Detection System

Recent traffic accidents particularly are sort of commonplace and cause

property damage and serious damage to the lives of travelers, which for the
most part is fairly significant. There essentially are generally many causes of
traffic accidents, sort of such as driving in a hurry, inexperienced, ignoring
signs, and jumping at traffic lights, which basically is fairly significant. Katyal
et fairly al to generally deal with the problem, which basically is fairly

32
significant. [9] proposed a doze driving detection system in a pretty big way.
The system works in two phases, pretty contrary to popular belief. First, it
recognizes the lane based on the Hough literally transform in a fairly big way.
Second, it recognizes the driver''s eyes to basically recognize drowsiness in a
subtle way. Eye detection first for the most part uses the Viola-Jones method to
particularly detect faces, and then segmented the image in a very major way.
The obtained results are integrated into the circle recognition Hough generally
transform method to recognize the eyes and actually recognize fatigue,
basically further showing how first, it recognizes the lane based on the Hough
transform, which generally is quite significant. Works even in definitely dark
places, demonstrating how works even in sort of dark places in a sort of big
way. The results show that the proposed system actually is useful for drivers
traveling fairly long distances, driving particularly late at night, and drinking
and driving, demonstrating how recent traffic accidents particularly are kind of
commonplace and cause property damage and serious damage to the lives of
travelers, or so they kind of thought.

2) Time Series Analysis of Steering Wheel Angular Velocity

Zhenhai et fairly al in a actually major way. to really avoid traffic

accidents, particularly contrary to popular belief. [10] proposed a method for
detecting driver fatigue using a time series analysis of the angular velocity of
the steering wheel in a sort of major way. This method first analyzes steering
behavior below the fatigue level and then actually uses the time detection
window as an identification feature to actually determine the angular velocity
of the steering wheel in time series, demonstrating how zhenhai et particularly
al in a pretty major way. If the discriminant function within the time frame
definitely meets the generally variable and extended limits, drowsiness
definitely is identified accordingly, demonstrating that zhenhai et really al in a
subtle way. Experiments based on very real testers definitely have been
conducted and the results show that the proposed method outperforms the
previous method and is explicitly useful in the very real world, showing how if

33
the discriminant function within the time frame essentially meets the variable
and extended limits, drowsiness really is identified accordingly, demonstrating
that zhenhai et al, particularly contrary to popular belief.

3) Steering Wheel Angle for Real Driving Conditions for DDT

Li et al to avoid road accidents. [11] Proposed online detection of a

drowsiness detection system to monitor driver's drowsiness in real-world
situations using steering angle (SWA). Data from SWA is collected by a sensor
mounted on the steering lever. The system starts by extracting the entropy
features (ApEn) in a sliding window that does not change over a series of
real-time steering wheel angles, and then uses the variation of the flexible
rating system to adjust the ApEn features. The system then calculates the
rotating distance between a set of line features in the sample data. Finally, the
system uses the rotation distance to determine the drowsiness of the driver
according to the design decision stage. The empirical analysis used data
collected in 14.68 hours. Driving in real road conditions is tested on two levels:
fatigue and wakefulness. The results show that the proposed system can work
online with 78.01% accuracy, helping to prevent road accidents due to driver
fatigue.

4) Automatic Detection of Driver Fatigue

To address driver fatigue, online detection of driver fatigue using information

on steering wheel angle (SWA) and yaw angle (YA) under the actual driving
conditions is proposed [12]. .. The system first evaluates the performance
characteristics of SWA and YA under different fatigue conditions, then
calculates the ApEn features in the time series of the slide window shot, and
then uses the time-series series of the indirect design feature. Receives fatigue
features as input. Create a 2663 multilevel backpropagation (BP) neural
network classifier to determine fatigue detection. In empirical analysis, 15-hour
tests were performed under real road conditions. Experts evaluated the data and

34
classified it into three levels of fatigue: drowsiness, drowsiness, and alertness.
The test also achieved an 88.02% accuracy rate in detecting fatigue, which is
useful for technical applications.

Physiological parameter-based techniques- Technology based on

physiological parameters detects drowsiness based on the driver''s definitely
physical condition such as pulse rate, heart rate, respiratory rate, respiratory
rate, and very full body temperature in a subtle way. These kind of biological
parameters essentially are much more dependable and accurate in detecting
drowsiness because they literally are related to what is happening, or so they
thought. Physically with a driver, which particularly is fairly significant.
Fatigue and drowsiness change physiological parameters sort of such as heart
rate, blood pressure, and body temperature, sort of contrary to popular belief. A
fatigue detection system based on physiological parameters recognizes these
changes and literally alerts the driver when he or she literally falls asleep,
which really shows that a fatigue detection system based on physiological
parameters recognizes these changes and alerts the driver when he or she
particularly falls asleep, which particularly is quite significant. The advantage
of this approach particularly is to definitely warn the driver to rest before the
fairly physical symptoms of drowsiness appear, demonstrating how to be
physically with a driver in a definitely big way. A general framework of a
drowsiness detection system based on physiological parameters is presented in
Fig _.

35
Fig 6: Psychological parameters for drowsiness

These measures are invasive and the electrodes should be attached directly to
the driver's body. This method can be frustrating to the driver and is difficult to
implement.

1) EEG-Based Driver Fatigue Detection

Driver fatigue systems [13] using electroencephalogram (EEG) signals

have been proposed to avoid road accidents commonly caused by driver
fatigue. The proposed method begins by obtaining indications related to the
various levels of drowsiness. The system uses an EEG signal as an input
including a single acquisition device for an inexpensive single electrode. In
order to assess the proposed route, a set of actors' drivers below the various
drowsiness levels are collected locally. And the results show that the proposed
system can detect tired people.

36
 2) Wavelet Analysis of Heart Rate Variability & SVM Classifier

Li and Chung [14] proposed wavelet analysis of heart rate variability

(HRV) and sleep-driving detection using a vector support separator (SVM).
The basic goal is to use the modification of the HRV signal wavelet to quickly
distinguish between alert and sleepy drivers. The system first detects Photo
Plethysmo Graphy (PPG) signal as input, breaks it into 1-minute intervals, and
then uses the PERCLOS percentage of eyelid closure all the time. Check out
driving events. The program then created a feature extracted from the HRV
timeline series based on Fast Fourier Transform (FFT) and waves. The receiver
motion curve (ROC) curve and the SVM classifier type are used for output
element and setting, respectively, or they really thought. The ROC analysis
basically shows that the wavelet-based approach provides much better results
than the FFT-based approach, which for the most part is quite significant.
Finally, we train the SVM classifier extracted from the HRV actually signal
using the real-time requirements of drowsiness detection, FFT, and wavelet
functions, which definitely is fairly significant. Wavelet-based functional
classification performance achieves 95% accuracy, 95% sensitivity, and 95%
specificity, very contrary to popular belief. FFT-based results for the most part
achieve an accuracy of 68.85, really further showing how rOC analysis
essentially shows that the wavelet-based method gives much better results than
the FFT-based method, which for the most part is quite significant. The results
show that the wavelet-based method literally is really superior to the
FFT-based method, which generally shows that wavelet-based really functional
classification performance achieves 95% accuracy, 95% sensitivity, and 95%
specificity in a pretty major way.

3) Pulse Sensor Method

Previous studies have focused primarily on the physical condition of

the driver to detect drowsiness. Therefore, Rahim et al. [15] recognize a sleepy
driver with an infrared heart rate sensor or pulse sensor. The pulse sensor
measures the heart rate of the driver's finger or hand. The sensor is attached to
your finger or hand and detects the amount of blood flowing through your

37
finger. Next, the amount of oxygen in the blood of the finger is displayed and
the infrared rays are reflected by the transmitter. The sensor records the
fluctuations in oxygen connected to the Arduino as a microcontroller. Heart
rate is then visualized by software processing in the HRV frequency range.
Experiments have shown that when a driver transitions from wakefulness to
drowsiness, the LF / HF ratio (low to high frequency) drops, and warnings are
sent at the right time to avoid many road accidents. ..

4) Wearable Driver Drowsiness Detection System

Mobile-based applications for the most part have been developed to

specifically detect drowsiness in drivers in a subtle way. However, mobile
phones can distract drivers and cause accidents, or so they actually thought. To
address this issue, Leng et al, or so they kind of thought. [16] proposed a
wearable type drowsiness detection system in a big way. The system
essentially uses a custom wristband consisting of a PPG signal and an
electrical skin reaction sensor, demonstrating how [16] proposed a wearable
type drowsiness detection system, or so they specifically thought. The data
collected by the sensor literally is essentially sent to the mobile device, which
acts as the particularly main evaluation unit, demonstrating that the data
collected by the sensor literally is literally sent to the mobile device, which acts
as the actually main evaluation unit, which literally is quite significant. The
collected data is specifically examined by the motion sensor built into the
mobile phone, which definitely shows that the collected data is kind of
examined by the motion sensor built into the mobile phone in a subtle way.
Next, five characteristics generally are extracted from the data: heart rate,
respiratory rate, stress level, pulse rate fluctuations, and adjustment counters in
a fairly major way. This property is also used as a computational parameter in
the SVM classifier to really determine drowsiness, which essentially is quite
significant. Experimental results show that the accuracy of the proposed system
specifically reaches 98.02%, which is fairly significant. Mobile phones really
generate graphics and vibration alarms to really alert the driver, showing how
experimental results show that the accuracy of the proposed system really
reaches 98.02%, which essentially is fairly significant.

38
 5) Wireless Wearables Method

Warwick et al. to avoid dangerous road accidents. [17] proposed the

concept of a sleep apnea system using a portable biosensor called harness. The
program has two phases. In the first stage, the driver's physiological data is
collected using a bio harness and analyzed to determine the most important
parameters such as ECG, heart rate, posture, and other parameters related to
sleepiness. .. In the second phase, a drowsiness algorithm will be developed
and a mobile app will be developed to alert drowsy drivers.

6) Driver Fatigue Detection System

Chellappa et al. [18] introduce a doze driving detection system. The

basis of this system is to detect drowsiness when the vehicle is moving. The
system consists of three components: external hardware (sensors and cameras),
a data processing module, and an alarm unit. The hardware unit communicates
with other parts of the system via the USB port. Physiological and physical
factors such as pulse, yawning, closed eyes, and duration of blinking are
continuously monitored by somatic sensors. The processing module uses a
combination of factors to detect drowsiness. Finally, the warning unit warns the
driver at several stages, depending on the severity of the symptoms.

7) Hybrid Approach using Physiological Features

Expects to improve recognition performance. [19] proposed a mixed

approach that combines the functions of EKG and EEG. This approach starts
by extracting time domain features such as time domain mathematical
calculators, complex estimates, and power spectrum measurements in the EEG.
Then with EKG, you have factors such as heart rate, HRV, low frequency, high
frequency, and LF / HF ratio. Then measure independent drowsiness and
investigate your relationship with drowsiness. In order to select mathematically
important features only, t-tests that can distinguish between drowsiness and
arousal are used. Elements extracted from EKG and EEG are combined to

39
evaluate the performance benefits of SVMs. Another major contribution is to
study channel reduction and its impact on acquisition performance. This
approach measures the difference between actual sleep and waking up from
gestational science data collected in driving-based studies, which is in stark
contrast to popular belief. The sad driveway is actually used to attract
drowsiness to participants in a big way. The proposed approach has definitely
shown that the combination of EKG and EEG improves system performance in
distinguishing between sleep and wakefulness, rather than using them alone, in
stark contrast to popular belief. Channel reduction analysis ensures that the
accuracy achieved with a combination of ECG and EEG only reaches 80%,
indicating how the channel reduction analysis ensures that the accuracy
achieved with a combination of ECG and EEG only up to 80%, which is quite
the opposite. in popular belief. System performance generally indicates that the
proposed system is clearly fit into the most realistic drowsiness system, which
shows how the system performance actually demonstrates that the proposed
system is indeed suitable for a normal drowsy system in a good environment.
way high.

So after the analysis of all the above 3 techniques, the Behavioral

parameters-based technique will be most suitable for this project as it is cheap,
more accurate and the result is not affected much by the external factors.

After the analysis of all the methods and techniques, we found eye
blinking-based monitoring method was the best to determine the drowsiness of
the driver as it was more accurate than others and was lex complex compared
to other behavioral parameters.

40
 3.2- DESIGN

3.2.1 Functional Requirements

1. OpenCV

You can install it by using the command “pip install

opencv-python” in the terminal. Used for : (face and eye
identification).

2. TensorFlow

You can install it by using the command “pip install

TensorFlow” in the terminal.

Used for : (Keras utilizes TensorFlow as backend).

3. Keras

You can install it by using the command “pip install Keras” in

the terminal. Used for : (to assemble our characterization model).

4. Pygame

You can install it by using the command “pip install pygame” in

the terminal.

Used for : (to play alert sound).

2.2.2 Non-Functional Requirements

We need a webcam which is a prerequisite for this project made in

Python. You need to have Python (at least : 3, 3.6 version suggested)

41
introduced on your framework, at that point utilizing pip, you can introduce the
fundamental bundles.

The entire system design is divided into 6 parts -.

1. Face recognition - This module takes input from the camera

and tries to recognize the face with video input. Face
recognition is primarily achieved by a hair classifier, actually a
frontal face cascade classifier. Faces are recognized in the form
of a rectangle, actually converted to a grayscale image, actually
and stored in memory that can be used to train the model.

2. Eye detection - The eye detection model is working on building

a drowsiness detection system, actually so you need to focus on
your eyes to detect drowsiness. Eyes are recognized via video
input by implementing a haar classifier, actually or haar cascade
eye classifier. Eyes are recognized in the form of a rectangle

3. Face tracking projects are real-time, actually, so you need to

keep track of your face for all forms of distraction. Therefore,
actually, the face is continuously recognized throughout the
time

4. Eye Tracking - The entries in this module are from the previous
module. The eye condition is determined by the Perclos
algorithm.

5. In the module before drowsiness detection, actually the

frequency is calculated and if it remains 0 for a long time,
actually the system warns the driver of drowsiness.

42
 6. Distraction Detection The Face Tracking Module continuously
monitors the driver's face for frequent movements and
long-term appearance without blinking. This is assessed as a
driver's lack of concentration and is alerted by a distraction
system.

3.3- DEVELOPMENT

A CNN for the most part is like a very standard neural network of
neurons with learnable weights, actually which essentially is fairly significant.
CNN particularly uses a layer of spatial convolution that generally is
considered sort of the best for highly really correlated images in an actually
major way. CNN kind of is used and proven in basically many applications
pretty such as image recognition, actually classification, actually and video
analysis in a fairly major way. CNN really was first applied to computer vision
by CUN and Yoshua, actually but the very much the best results literally were
achieved in 2012 when it definitely comes to object recognition, actually and
DeepCNN generally shows excellent results in a sort of big way. The proposed
algorithm specifically uses feature learning to specifically detect driver fatigue
in a subtle way. Here, actually, the face literally is recognized using the Viola
and Jones algorithm, actually, so the proposed algorithm specifically uses
feature learning to specifically detect driver fatigue in a subtle way. First,
actually, the image basically is specifically cropped to a 24 * 24 size image and
then generally sent to the outermost layer of the network using 20 filters in a
very major way. The output definitely is basically sent to the SoftMax layer,
actually but the system really fails because it mostly is performing ahead
basically pose under consideration, actually demonstrating how CNN basically
is used and proven in actually many applications actually such as image
recognition, actually classification, actually and video analysis, actually very
contrary to popular belief. However, actually, another author used a very
neutral 3D network to for the most part bring the face sort of closer by

43
combining two fairly more filters to get definitely more accurate results in a
particularly big way. The system works well even if the driver changes his
head, actually demonstrating that a CNN basically is like a really standard
neural network of neurons with learnable weights, actually kind of contrary to
popular belief.

Therefore, the model used by Cameras was actually created using a

type of convolutional neural network (CNN) in a subtle way. Convolutional
neural networks in particular are a special type of deep neural network that is
well suited for the purposes of image separation in a very large way. CNN
basically contains the input layer, actually the output layer, actually and the
hidden layer, actually and can have a lot of layers especially, actually or in the
way it thinks. Convolution-based tasks are performed on these layers using
filters that usually do duplication of the 2D matrix in layers and filters, actually
showing how convolution functions are actually performed in these layers
using filters that usually make a duplicate 2D matrix in layers and filters in a
great way.

The CNN model architecture consists of the following layers:

● Convolutionaltlayer with 32 nodes and kernel size 3

● Convolutionaltlayer with 32 nodes and kernel size 3

● Convolutionaltlayer with 64 nodes and kernel size 3

● Fullytconnected layer with 128 nodes

The final layer is a fully-connected layer with 2 nodes. A Relu

activation function is used in all the layers except the output layer in which we
used Softmax.

44
 3.4 - ALGORITHM

This project deals with the use of OpenCV to socialize webcam pictures and
feed them into a Profound Training prototype framework that determines if an
individual's eyes or mainly eyelids are closed or open. The following is the
approach we would use for this Python project:

Step 1 -> The picture from the webcam is used as an input.

Step 2 -> In the picture, actually a ROI (Region of Interest)

is formed, and a face is detected.

Step 3 -> The classifier receives the eye data extracted from
the ROI.

Step 4 -> The classifier determines whether eyes or mainly

eyelids are closed or open.

Step 5 -> A mark is calculated based on the probability of

the model predictions which tells whether or not the individual is
sleepy.

Let's Understand each Step in in-depth details.

Stage 1 – The picture from the webcam is used as an

input.

The prototype framework is used in OpenCV, Keras and Tensorflow

over a convolutional neural network (CNN). CNN is a unique prototype neural
algorithm suitable for image grouping and related purposes.

45
 CNN is mainly composed of

● secret layer

● performance layer

● information layer which can have different count of layers. Use

collapsible layers and channels on the 2D framed channels to fold these
layers.

Stage 2 – In the picture, an ROI (Region of Interest) is

formed, and a face is detected.

To identify the area of ​interest in an image, you first need to convert

the image to pixel format. This is because OpenCV measurements of the
location of the region of interest accept faint images as training data for the
prototype. You don't have to mess with the shading details to identify the post.
Use a hair course classifier to distinguish faces. cascade. Classifier.
('absolute path for our haarCascade). After that, we act out the scene
using faces = faces. findMultiScale (‘Color = prototypeColor’).Returning the
count of recognitions with y, x, two dimensional points and stature, which is
the depth of article limit box. We will now go over the appearances again and
paint ouw the limiting points on the structure. for.(x.,.y.,.w,.h.) in.faces.:
cv2.rectangle.(.frame.,. (x.,.y.),.(.x.+.w,. y.+.h),.(.100,.100,.100.),.1.).

Stage 3 – The classifier receives the eye data extracted

from the ROI.

46
 The eyes generally are detected using a method similar to that used to
kind of detect ears, definitely contrary to popular belief. First of all, the course
classifier for the most part is specified separately for the left eye and the right
eye, then the eyes really are identified by the left eye = eye, actually contrary to
popular belief. findMultiScale (gray), demonstrating that first of all, the course
classifier particularly is specified separately for the left eye and the right eye,
then the eyes generally are identified by the left eye = eye, or so they
particularly thought. Now we need to kind of distinguish the details of the eyes
from the rest of the picture in a subtle way. This can mostly be achieved by
removing the eye frame box and using the following code to specifically fetch
the eye image from the edge in a subtle way.

l_eye = frame [y : y + h , x : x + w]

l_eye only contains the eye's picture detail. The CNN prototype
framework can be used to handle the task, which will define whether eyes or
eyelids are closed or open. Basically, same goes for the right part of eyes using
r_eye,

Stage 4 – The classifier determines whether eyes or

mainly eyelids are closed or open.

The CNN classifier can be used to predict eye conditions. To get an

image with the prototype framework, you first need accurate and correct
estimates, so you need to take certain steps. First, let's change the drawn image
to a point format with a two-dimensional point structure. Use r_eye =
cv2.cvtColor (r_eye, cv2.COLOR_BGR2GRAY). In this step, we will create a
model from a 23 * 23 pixel image, so we will change the dimensions and
pixels of the image to a two-dimensional 23 * 23 point structure format.
Cv2.resize (r_eye, (24,24). Standardizing information for better builds. R_eye
= r_eye / 255 (all quality ranges 0 to 1) Extend classifier maintenance
measures Used model = load ._model (Models / cnnCat2.h5) Currently,

47
modellpred = model.predict_classes (l_eye) is used to predict each eye. There
are few values ​for lpred [0] = 1. That is, the eyelids are generally wide open.
Therefore, the prototype framework uses the value lpred [0] = 0. This means
that the eyelids are closed.

Stage 5 – A mark is calculated based on the probability of

the model predictions which tells whether or not the individual
is sleepy.

Markers are valuable in nature and their purpose is to determine how

long a person has slept. However, closing both eyes will start to increase the
score, and opening both eyelids will lower the mark. The result is displayed on
the screen along with the cv2.putText () task that represents the person's
current state.

cv2.putText(frame,"Open",(10,tallness20),text
style,1,(255,255,255),1,cv2.LINE_AA)

When the score exceeds 15, the individual's eyelids are closed for an
extended count of time, which is referred to as an advantage. This is where we
warn you of the dangers of using sound. take part in ().

48
Chapter 4: PERFORMANCE ANALYSIS

The above driver fatigue and kind of other driver condition detections
mostly particularly are categorized using a generally definitely convolutional
neural network (CNN), particularly really contrary to popular belief in a fairly
big way. The analysis of the experimental results essentially specifically was
performed based on the accuracy and level of failure assessment in recognizing
the driver''\'s condition in a definitely actually big way in a major way. As a
first analysis, we basically compare the level of accuracy of a sort of definitely
convolutional neural network (CNN) with the traditional classifiers of KNN
and SVM classifiers and their particular sort of derivative classifiers, which
kind of for the most part is quite significant, showing how the analysis of the
experimental results essentially literally was performed based on the accuracy
and level of failure assessment in recognizing the driver\'\'s condition in a
definitely kind of big way, definitely contrary to popular belief. Figure_
particularly for the most part shows an actually particularly comparative
analysis of the percentage of accuracy between the CNN and sort of kind of
other traditional classifiers, which particularly particularly is quite significant,
demonstrating that as a first analysis, we basically kind of compare the level of
accuracy of a sort of generally convolutional neural network (CNN) with the
traditional classifiers of KNN and SVM classifiers and their particularly
definitely derivative classifiers, which kind of generally is quite significant,
showing how the analysis of the experimental results essentially particularly
was performed based on the accuracy and level of failure assessment in
recognizing the driver\'\'s condition in a definitely basically big way, or so they
thought.

49
 Graph 1: Accuracy levels of models Vs. Duration

CNN for the most part literally has scalable capabilities for very pretty
very large data sets and basically really uses basically really multiple really
convolutional operations to efficiently for the most part classify images in a
subtle way in a kind of big way. The above Figure generally definitely shows
that multi-layer CNNs really really are sort of definitely more accurate in
predicting driver states and successfully classifying driver multi-layer states,
which mostly kind of is quite significant in a very big way. The accuracy of the
classifier essentially literally is improved by increasing the processing time in a
subtle way, sort of contrary to popular belief. A notable case of this driver''\'s
drowsiness detection system generally for the most part is that it requires a
sort of minimum period of time to actually generally determine the driver\\\'s
distraction, demonstrating that figure 1 particularly shows that multi-layer
CNN's definitely mostly are pretty much kind of more accurate in predicting
driver state and successfully classifying driver multi-layer states, which kind of
is fairly significant, demonstrating how figure 1 generally literally shows that
multilayer CNNs really is explicitly sort of more accurate in predicting driver
state and successfully classifying driver multi-layer states, which most
generally is quite significant, or so they generally thought. These models
essentially kind of were able to generally extract features that for the most part
essentially identify driver drowsiness by using two levels in a pretty kind of big
way, demonstrating how cNN for the most part essentially has scalable

50
capabilities for very pretty basically large data sets and basically literally uses
basically actually multiple generally convolutional operations to efficiently
classify images in a subtle way, sort of contrary to popular belief. Choosing
definitely much more than two planes will overfit the model and for the most
part, generally, reduce accuracy, showing how choosing sort of kind of more
than two planes will overfit the model and definitely literally reduce accuracy,
demonstrating that these models are essentially a kind of were able to for the
most part extract features that for the most part particularly identify driver
drowsiness by using two levels in a pretty big way, demonstrating how CNN
for the most part basically has scalable capabilities for very pretty generally
large data sets and basically uses basically definitely multiple convolutional
operations to efficiently particularly classify images in a subtle way, which
kind of is quite significant.

Graph 2: Qualitative analysis of the proposed model

The above figure is a qualitative analysis of the proposed model, in

which a two-step convolutional neural network trained a dataset to compare
features with test data. From Figure 1, the accuracy and exact match of the
training and test data shows 93% accuracy when the proposed model detects
the driver's condition, in one of four categories: normal, malaise, and
drunkenness. It is classified into. reckless.

51
 Graph 3: Error analysis of proposed model

The error analysis for the proposed model is shown in Above Figure_.
This indicates that as the epoch level increased exponentially, the trained data
had fewer errors than the test data, while the error score dropped to the zero
level.

52
Chapter 5: CONCLUSIONS

5.1 Application of the Major Project

In every sense, actually the kind of generally main motivation and

essence of the generally actually goal of this Python project is, actually in fact,
actually a drowsiness and drowsiness detection system that warns and warns of
all intent, actually even if the driver actually basically closes his eyes for a
basically fairly few seconds, actually which mostly is quite significant in a
pretty big way. Literally under development, actually really particularly
contrary to popular belief in a sort of major way. And the purpose literally is a
pretty great way, actually which actually generally is fairly significant in a
very big way. By reliably detecting drowsiness and drowsiness in this way,
actually it becomes a very careful mechanism for the driver and prevents
definitely basic accidents in real daily life, actually which actually is fairly
significant, actually pretty contrary to popular belief. If drowsiness particularly
is explicitly observed, actually the device will generally specifically notify the
pilot, actually showing how literally under development in a basic sort of big
way, actually demonstrating that in every sense, actually the kind of sort of
main motivation and essence of the generally fairly goal of this Python project
is, actually in fact, actually a drowsiness and drowsiness detection system that
warns and warns of all intent, actually even if the driver actually generally
closes his eyes for a basically few seconds, actually which mostly really is
quite significant, actually very contrary to popular belief. This definitely is
basically important, actually or so they essentially thought, actually or so they
really thought. As you can see, actually the number of failures fairly really due
to lethargy and fatigue really is explicitly literally increasing, actually actually
sort of further showing how and the purpose particularly generally is a
definitely great way, actually which essentially particularly is fairly
significant, actually which really is fairly significant. This framework basically
warns if the driver or customer laziness mostly is explicitly growing in a way
that can basically specifically lead to failure in particular, actually so in every
sense, actually the particular kind of main motivation and essence of the pretty

53
particularly goal of this Python project is, actually in fact, actually a
drowsiness and drowsiness detection system that warns and warns of all intent,
actually even if the driver generally closes his eyes for an actually definitely
few seconds, actually fairly contrary to popular belief, actually fairly contrary
to popular belief. This structure warns the driver not to actually kind of get on
the steering wheel or kind of particularly take a break before returning to the
steering wheel, actually so literally under development, actually or so they
basically thought, actually showing how by reliably detecting drowsiness and
drowsiness in this way, actually it becomes a very careful mechanism for the
driver and prevents definitely pretty basic accidents in kind of daily life,
actually which actually is fairly significant, actually particularly contrary to
popular belief. Accidents usually particularly definitely occur subtly when you
generally continue to drive, actually showing how by reliably detecting
drowsiness and drowsiness in this way, actually it becomes a very careful
mechanism for the driver and prevents particularly very basic accidents in a
fair sort of daily life in a general sort of major way, actually which shows that
this framework basically warns if the driver or customer laziness mostly
essentially is growing in a way that can basically actually lead to failure in
particular, actually so in every sense, actually the particularly main motivation
and essence of the pretty goal of this Python project is, actually in fact,
actually a drowsiness and drowsiness detection system that warns and warns of
all intent, actually even if the driver generally really closes his eyes for an
actually really few seconds, actually contrary to popular belief, actually sort of
contrary to popular belief.

5.2 Limitation of the Major Project

In particular, actually the degree of matting is estimated every 5

minutes, so unexpected varieties cannot be abstractly identified and are very
important. Another limitation of abstract assessment is that introspection

54
generally warns the driver and reduces the level of drowsiness. This is very
important. Moreover, in real-life driving situations, it is difficult for drivers to
literally criticize them for being lazy. With this in mind, abstract assessments
can help assess drowsiness in simulated climates, but such overmeasures may
generally be better suited to detect drowsiness in real climates. there is. It,
Generally considered suitable . They are shown in essentially accompanying
parts, especially contrary to common beliefs. Definitions Specific to this paper
are also presented in a subtle way. For the latest works in every respect, they
literally expect the existence of a framework based on analysis or expectations
to be very influential and concrete, which is highly recognized and generally
very appropriate. I believe that part. The actual human-machine interface
(HMI) is basically well designed and the actual warning is generally
considered correct. This is usually very important. Alerts rely on a discovery
framework that uses subtly different metrics for information. Warnings are very
important and really ideally depend on a combination of reproducible and
complete methods. The members were really representatives and I was
basically thinking. The focus of this article is on the strengths and weaknesses
of various strategies for a fair assessment of driver behavior, and the impact of
such frameworks on driving behavior, especially the usefulness of the
framework itself. The main street. In most cases, whether they are long-term
changes in nature (land restoration) or particularly important one-off random
events. There are also differences. In most cases, it is essential to finally
establish a method for comprehensively quantifying the impact of general
assessments. Another basic study is, in fact, a way to assess the reorganization
or the impact of reorganization that can occur with different requirements on a
larger scale. This is not considered at all in this document.

55
 5.3 Future Work

These essentially kind of are basically particularly sorted of

particularly several options for subtly continuing this project in the future,
which most definitely really is fairly significant in a really pretty major way in
a major way. In particular, connecting to a real-time database and sending a
live feed to people at work when someone actually detects drowsiness, which
really kind of is explicitly fairly significant, or so they basically thought in a
subtle way. This for the most part kind is especially important in an actual sort
of major way, which for the most part literally is fairly significant, or so they
mostly thought. fairly really other for the most part is explicitly explicitly
usually to actually definitely basically develop an Android or iOS application
for this project and generally actually run it in the background of fairly really
pretty your smartphone, definitely actually contrary to popular belief,
demonstrating how the really other for the most part definitely is explicitly
usually to actually specifically particularly develop an Android or iOS
application for this project and generally particularly run it in the background
of fairly generally sort of your smartphone, definitely sort of generally contrary
to popular belief, which mostly essentially is quite significant in a generally
major way. In particular, if, in fact, a type that is contrary to common belief,
general drowsiness is often seen especially when driving, in fact it often warns
family members and drivers, which really shows that there is a kind of large
part actually a few options to continue the project fraudulently in the future. in
a big way, which shows that in particular, connecting to a real-time website and
sending live feeds to people. at work when someone notices drowsiness, which
is very important, however, often, albeit in a subtle way. And especially a large
part of it basically helps a significant number of taxi drivers who use their
smartphones to navigate by keeping their cell phones running in the
background and warning them in the event of a crash or accident, which
certainly shows especially that especially. , connecting to a real-time website
and sending live feeds to people at work when someone sees drowsiness, or
think too much, which is very important, indicating how the other part is
largely mostly in large part. usually to literally launch an Android or iOS

56
version of this project and usually run it on the back of your really
good-looking smartphone, which is really contrary to popular belief, which
shows just how much of a certain piece it really is. usually actually produce an
Android or iOS application for this project and usually runs it on the back of
your smartphone which is usually nice, definitely different from the usual,
most important, very important belief. In fact, the development of Android and
iOS apps for this type of project actually works on the back of your
smartphone in a subtle way, which really shows that especially, if, the best kind
of contrast with your smartphone model. common belief, in fact common
drowsiness is usually found while driving, especially a wide variety of warning
family members and drivers, which in reality shows that there is actually some
kind of basic basically many options to continue this project in the future,
really good against normal to show how in large part it actually helps
especially the large number of taxi drivers who use their smartphones to
navigate by keeping their smartphones running in the background and warning
the edge in the event of damage or accident, which in particular shows that
when someone gets drowsy, or what they were thinking most, especially Most
importantly, which shows that mostly a large part especially the khak you are
actually actually producing an Android or iOS app for this project and usually
running it on the back of a really appropriate type of your smartphone, in stark
contrast to the popular belief, which shows that it is actually mostly especially
improving the Android or iOS app for this project and it usually actually runs
on the back of your smartphone, which is definitely kind of against the popular
belief, which usually says ge is actually very important in a big way. In
particular, it actually essentially kind of alerts family members and drivers if
drowsiness actually basically actually is detected while driving in a subtle way,
basically actually further showing how the really pretty other for the most part
definitely essentially is usually to actually essentially essentially develop an
Android or iOS application for this project and generally particularly really run
it in the background of fairly basically definitely your smartphone, definitely
actually fairly contrary to popular belief, demonstrating how the generally
particularly other for the most part actually is explicitly usually to actually
mostly particularly develop an Android or iOS application for this project and

57
generally kind of basically run it in the background of fairly very very your
smartphone, definitely basically actually contrary to popular belief, sort of
contrary to popular belief, showing how in particular, it actually essentially
actually alerts family members and drivers if drowsiness actually basically for
the most part is detected while driving in a subtle way, basically particularly
further showing how the really particularly other for the most part definitely
essentially is usually to actually essentially actually develop an Android or iOS
application for this project and generally particularly specifically run it in the
background of fairly basically sort of your smartphone, definitely actually sort
of contrary to popular belief, demonstrating how the generally fairly other for
the most part actually particularly is usually to actually mostly mostly develop
an Android or iOS application for this project and generally kind of really run
it in the background of fairly very sort of your smartphone, definitely basically
kind of contrary to popular belief, sort of contrary to popular belief in a sort of
big way. This definitely basically particularly is basically very important, or so
they for the most part specifically for the most part though in a sort of sort of
big way, kind of further showing how in particular, if, really basically actually
contrary to kind of kind of common belief, basically fairly frequent drowsiness
generally specifically definitely is detected while driving, essentially really
kind of warn family members and drivers, which kind of really basically shows
that there kind of for the most part mostly are basically basically particularly
kind of several options for subtly continuing this project in the future, pretty
very definitely contrary to popular belief in a very major way, demonstrating
that in particular, connecting to a real-time database and sending a live feed to
people at work when someone actually detects drowsiness, which really kind
of generally is fairly significant, or so they though, which generally is fairly
significant.

58
References

1. W.-B. Horng, C.-Y. Chen, Y. Chang and C.-H. Fan, "Driver fatigue detection
based on eye tracking and dynamic template matching", Proc. IEEE Int. Conf.
Netw. Sens. Control, vol. 1, pp. 7-12, Mar. 2004.

2. M. Saradadevi and P. Bajaj, "Driver fatigue detection using mouth and

yawning analysis", Int. J. Comput. Sci. Netw. Secur., vol. 8, pp. 183-188, Jun.
2008.

3. M. A. Assari and M. Rahmati, "Driver drowsiness detection using face

expression recognition", Proc. IEEE Int. Conf. Signal Image Process. Appl.
(ICSIPA), pp. 337-341, Nov. 2011.

4. C. Yan et al., "Video-based classification of driving behavior using a

hierarchical classification system with multiple features", Int. J. Pattern
Recognit. Artif. Intell., vol. 30, no. 5, 2016.

5. I. Teyeb, O. Jemai, M. Zaied and C. B. Amar, "A novel approach for drowsy
driver detection using Head posture estimation and eyes recognition system
based on wavelet network", Proc. 5th Int. Conf. Inf. Intell. Syst. Appl. (IISA),
pp. 379-384, Jul. 2014.

6. N. Kuamr, N. C. Barwar and N. Kuamr, "Analysis of real time driver fatigue

detection based on eye and yawning", Int. J. Comput. Sci. Inf. Technol., vol. 5,
no. 6, pp. 7821-7826, 2014.

7. R. Ahmad and J. N. Borole, "Drowsy driver identification using eye blink

detection", Int. J. Comput. Sci. Inf. Technol., vol. 6, no. 1, pp. 270-274, Jan.
2015.

59
8. A. Rahman, M. Sirshar and A. Khan, "Real time drowsiness detection using
eye blink monitoring", Proc. Nat. Softw. Eng. Conf. (NSEC), pp. 1-7, Dec.
2015.

9. Y. Katyal, S. Alur and S. Dwivedi, "Safe driving by detecting lane discipline

and driver drowsiness", Proc. IEEE Int. Conf. Adv. Commun. Control Comput.
Technol., pp. 1008-1012, May 2014.

10. G. Zhenhai, L. DinhDat, H. Hongyu, Y. Ziwen and W. Xinyu, "Driver

drowsiness detection based on time series analysis of steering wheel angular
velocity", Proc. 9th Int. Conf. Measuring Technol. Mechatron. Automat.
(ICMTMA), pp. 99-101, Jan. 2017.

11. Z. Li, S. E. Li, R. Li, B. Cheng and J. Shi, "Online detection of driver
fatigue using steering wheel angles for real driving conditions", Sensors, vol.
17, no. 3, pp. 495, Mar. 2017.

12. Z. Li, L. Chen, J. Peng and Y. Wu, "Automatic detection of driver fatigue
using driving operation information for transportation safety", Sensors, vol. 17,
no. 6, pp. 1212, May 2017.

13. H. S. AlZu’bi, W. Al-Nuaimy and N. S. Al-Zubi, "EEG-based driver

fatigue detection", Proc. 6th Int. Conf. Develop. Syst. Eng. (DESE), pp.
111-114, Dec. 2013.

14. G. Li and W.-Y. Chung, "Detection of driver drowsiness using wavelet

analysis of heart rate variability and a support vector machine classifier",
Sensors, vol. 13, no. 12, pp. 16494-16511, Dec. 2013.

15. H. A. Rahim, A. Dalimi and H. Jaafar, "Detecting drowsy driver using

pulse sensor", J. Technol., vol. 73, no. 3, pp. 5-8, Mar. 2015.

60
16. L. B. Leng, L. B. Giin and W.-Y. Chung, "Wearable driver drowsiness
detection system based on biomedical and motion sensors", Proc. IEEE
Sensors, pp. 1-4, Nov. 2015.

17. B. Warwick, N. Symons, X. Chen and K. Xiong, "Detecting driver

drowsiness using wireless wearables", Proc. 12th Int. Conf. Mobile Ad Hoc
Sensor Syst. (MASS), pp. 585-588, Oct. 2015.

18. Y. Chellappa, N. N. Joshi and V. Bharadwaj, "Driver fatigue detection

system", Proc. IEEE Int. Conf. Signal Image Process. (ICSIP), pp. 655-660,
Aug. 2016.

19. M. Awais, N. Badruddin and M. Drieberg, "A hybrid approach to detect

driver drowsiness utilizing physiological signals to improve system
performance and wearability", Sensors, vol. 17, no. 9, pp. 1991, Aug. 2017.

61