2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N)

Driver Drowsiness Detection using Deep

2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N) | 979-8-3503-3086-1/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICAC3N60023.2023.10541587

Learning; Approach towards Automating Object

Recognition
Dr. Jaya Sinha Pratham Mangla Sankalp Purwar
Computer Science and Engineering, Computer Science and Engineering, Computer Science and Engineering,
Galgotias College of Engineering and Galgotias College of Engineering and Galgotias College of Engineering and
Technology, Technology, Technology,
Greater Noida, India Greater Noida, India Greater Noida, India
[email protected] [email protected] [email protected]

Naman T. Akilan
Computer Science and Engineering Department of IQAC,
Galgotias College of Engineering and Amity University, Kolkata
Technology, [email protected]
Greater Noida, India
[email protected]

Abstract— Drowsiness is the state of being so tired that one constraints placed on the module given in this work is to
feels the urge to rest, and it can result in symptoms that prevent false alarms from activating, which could distract the
significantly affect how well tasks are performed, such as driver and force him or her to deactivate the advanced driver
sluggish response times, brief periods of not being conscious, or
assistance systems. This is because the advanced driver
microsleeps (blinks lasting more than 500 milliseconds), to
assistance systems will have multiple features incorporated.
mention a few. It is one of the major factors in either serious or
deadly car accidents. These symptoms are highly dangerous The experimental methodology used in the majority of the
when driving because they greatly increase the likelihood that published works involves removing and classifying individual
drivers will miss road signs or exits, drift into other lanes, or frames from each video and then determining whether the
even crash their car and cause an accident. Drivers who are classification is accurate or not. However, this methodology
drowsy crash their cars more frequently than those who are ignores the inherent relationship between successive images,
alert. So, to avoid these kinds of situations a safety system is used making false positive measurements less reliable. There aren't
that can help reducing fatal accidents by detecting whether the many studies out there right now that test the systems on
driver is in drowsy state or not. If the driver is found in drowsy
whole videos and track how many alerts are triggered during
state, the system will start to beep. Hence alerting the driver and
each video (which is an important aspect when evaluating the
co-passengers about the situation. So, the overall goal of our
project named driver drowsiness detection is to identify the state number of false alarms raised during a period of time). Thus
when driver feel drowsy and alert him so that fatalities our project aims to device an easy technique which can be
can be terminated. used to detect tiredness levels of driver while operating his
vehicle. We used a CNN architecture to build the drowsiness
Keywords—Drowsiness, Deep Learning, CNN, Microsleep detecting application. In our research, we have developed a
technique by using deep learning and deviced a model which
I. INTRODUCTION can detect facial features of driver using that finds the state of
Driver Drowsiness is one among many reasons which eyes as closed or open.
leads to fatal damages and even deaths during car accidents. A. Different approaches of face recognition:
According to prior studies, 10–30% of car accidents are the
result of drowsiness or driver weariness, which are primarily Depending on the area of analysis, several methods for
brought on by lack of sleep, intoxication, drug use, exposure detecting driver intoxication could be categorized. the first
to the sun, or alcohol and/or alcohol. These symptoms are technique involves analyzing the driver's activity to find any
highly dangerous when driving because they greatly raise the modification or suspicion in it. The technique uses visual
chances that drivers will crash their car, miss exits or road sensors to analyze face features like the lips and eyes, or it
signs, drift into other lanes, or possibly miss them altogether. looks for particular patterns in electrophysiological impulses
The risk of an automobile accident is higher for sleepy drivers that happen when a driver nods off. Some methods analyze
than for alert people. The technology can alert the driver and signals taken from the steering system in order to deduce a
prevent accidents by using that information. One of the driver's state indirectly. According to some sources, the

ISBN: 979-8-3503-3086-1/23/$31.00©2023 IEEE 641

Authorized licensed use limited to: AMITY University. Downloaded on June 06,2024 at 12:34:24 UTC from IEEE Xplore. Restrictions apply.
 2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N)

PERCLOS strategy had a success rate of more than 90% engine fires up. This primarily enables us to continuously
[1][2][3]. monitor and record the driver's eye blinking [9].
An approach that makes use of a information based on Deng et al. evaluate drowsiness using the closer eye and
transform packet model has been presented to spot the driver's mouth angle. The method also installs a pre processing stage
tiredness. Factors related to them led to approaches based on in it to find very low illumination conditions by self and
biological signals collected from drivers producing findings enhance the quality of image whenever required [10].
for detection that were more precise than those relying on the
Because sleep-related crashes make up a significant portion of
signals coming from the response on vehicle's steer wheel. A
auto accidents globally, researchers and automakers have
fundamental drawback in this type is that it requires to attach
come up with a number of valid ideas, with the help of patterns
physiological signal-sensing apparatus to the driver, including
and ways in driving behavior to tracking the vehicle driver's
EEG, ECG, and EOG [4][5].
necessary signs and driver's brain activities when drivers are
Other than identifying and processing biomedical signals, operating a vehicle. The next paragraphs provide an overview
there are some other techniques which uses face analysis for of the three groups into which the majority of them fall.
extracting different features like eye area of the driver or Finding variations in the conduct of the vehicle is one tactic,
his\her mouth or other region of interests through computer as suggested by McDonald. He developed a theoretical
vision. A technique having some limited steps can also be used method that is dependent on time and that makes use of the
which recognizes face and eyes and tracks them, after which speed of driver's vehicle and accelerator pedal position (APP)
it analyses the situation to maintain performance. In varied and also use the vehicle's steering angle. That quantities are
lighting situations, the system filters images for the eye. This fed into the present networks like Bayesian network, that helps
programme evaluates the closure. The PERCOLS to assess whether the driver exhibits symptoms of
measurement is used to evaluate the state of the eye [6][7]. intoxication. It was discovered that the algorithm had lesser
false positive values and rates than the present and evolved
B. Motivation
PERCLOS approaches, that determine tiredness depending
Our project will address the huge problem of driver upon patterns of the eyelids and movements [11].
drowsiness. Most accidents today occur as a result of driver
This is very crucial for examining the information that the
drowsiness, which is a very significant problem. This
method has gathered over the earlier 10 seconds time interval
condition is used to construct a deep learning system for driver
in order to ascertain whether a person is at risk of lane
drowsiness detection.
departures brought on by tiredness. A second strategy is based
C. Novelty of Idea on research that aims to forecast crashes by monitoring the
drivers' vital signs, brain waves, and electroencephalogram
Our idea for this work is that a camera installed in a car will
(EEG) measurements. Wei et al. contrasted the less
record frontal photos of the driver, which will further then be
pleasurable whole scalp EEGs used in laboratories with the
analysed using artificial intelligence (AI) techniques,
less intrusive and easier-to-wear EEG Brain Computer
including deep learning, for determining whether the vehicle
Interfaces. This breakthrough opens the door to the
driver is sleepy or not.
development of more aesthetically pleasing and unobtrusive
head bands. Because only EEG is not able to find and analyze
II. LITERATURE SURVEY
all the forms of driver drowsiness. Then, Kartsch et al.
By combining several different sleepiness tests, Dasgupta generated the method by combining EEG with the current and
et al. increase the likelihood of accurately predicting drowsy highly advanced Inertial Measurements Units (IMU) sensors
condition. The verification of sleep is done in three steps. for detecting all five stages of the fatigue and giving an
These steps are totally depends upon inputs from sight, voice, accuracy of about 95%. The scientists merged EEG data with
and driver's touch. In the initial stage, a linear SVM is totally behavioural data from the IMU in order to identify tiredness.
used to properly differentiate and identify the eyes of the Another shortcoming of the EEG technology was its need on
driver as eyes are open or eyes are closed. This composition power. Their methods and algorithms also helped in the
method actually aids in the correctness of this model, which is installation of the current parallel ultra low power (PULP)
quite impressive. However, the time needed to complete the platforms on the recent microcontroller, that is helping in
tests suffers as a result. Our study predicts that, in the worst- enhancing the vehicle battery life installed in vehicle to about
case scenario, it will take the application 20 seconds after the 46 hours and helping in making the way for the
sleep onset to complete all of its tests and sound the alarm. development of systems that are almost always-on and have
The final level of verification also calls for the driver to very low system maintenance. Tateno et al. also developed a
manually interact, which might be distracting when they are algorithm that simply employs monitoring of real time heart
not sleepy [8]. rates for determining respiration of the driver in vehicle and
In his paper, S. Kailasam introduces a method that basically consequently, their level of alertness. The technique was
tracks the vehicle driver's face and eyes after the moment the found to be an accurate indication of respiration and,
consequently, drowsiness. An additional approach is to use

642

Authorized licensed use limited to: AMITY University. Downloaded on June 06,2024 at 12:34:24 UTC from IEEE Xplore. Restrictions apply.
 2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N)

computer vision as a tool. Computer vision now has additional the real time applications of the framework, an video input
detection and classification abilities because to recent was gathered by installing a good quality camera on the front
developments in deep learning. These described methods are side of the driver's vehicle. It is capable to accept the driver's
employed in vision-related programs in system for the hand's gestures, face patterns, and the upper body moves. Also
identification of objects, for condition of driver, and also for then, pre trained detector including landmarks of 68 facial
some other related purposes. The help provided by data having patterns from the recent library is utilized for real-time
large set of images had a major role in this field [12][13]. application and proper functioning of the model. A camera
that can fit the driver's face can be mounted on the car's
For determining the eyes of driver are shut or not, Tayab Khan
dashboard to collect the input footage. Implemented was a
and colleagues came up with an algorithm for measuring the
face detector that depends on the Oriented Gradients
angle of curvature of eyelid of driver's eye. They also got 95%
Histogram. Suggested approach taken the help of required
accuracy with this method, but it is seen that it contains a
ratios to track the blinking behavior of the driver [18].
limitation that for proper and accurate functioning of model it
requires proper light because model works poorly in dark light
and also at night. 350 pictures were trained using a customized III. PROBLEM FORMULATION AND METHODOLOGY
dataset by Shakeel et al. using the Mobile Net-SSD
A. Problem Statement
architecture. Approximate precision average of the model that
it could get or achieve was measured equal to 0.84. Because Electroencephalography (EEG) and electrocardiography
(ECG), which monitor heart rhythm and brain frequency,
the method or model developed can only work on the android
respectively, are two common drowsiness detection
device and smartphone and the stream of camera installed techniques that currently exist but are inconvenient to use
could be identified and mapped in real time, this making the while driving. These systems are not suitable for driving
system more economical for use and more effective for use. conditions. However, it is first necessary to identify the
The NTHU dataset was used for this investigation. The current physical cues that will indicate drowsiness in order to create a
advanced YawDD and current advanced NTHU-DDD sleepiness detection method that is reliable and accurate. The
databases were utilized in a different study by Xie et al. to employment of a camera placed directly in front of the driver
detect yawning using the recent learning techniques extracted as a sleepiness detection device is more appropriate. The
from yawning short video clips taken by installed device. The difficulties arise when finding a region of interest in the facial
system had the ability to measure faces more accurately and capture when light conditions are not appropriate or when
to be resistant to the differences occurring in the position of driver's face is at some angle. Apparently, the whole objective
face of driver and angle in regard to installed camera. is to evaluate the studies and ways about making the scheme
for detection of drowsiness that makes use of video or a
Accuracy of ability of android application developed by
webcam. It creates a system that can analyze each frame of the
Mehta et al. to compute the values of Eye Ratios based on
video after analyzing the recorded video images.
artificial intelligence (AI) models to predict drivers' level of
fatigue was 84%. Numerous companies are attempting to use B. Methodology
the strategy of combining the three aforementioned
Using Keras and Convolutional Neural Networks, we created
approaches and using a wide range of inputs to choose. Using
the model. The area of study and the technique in deep
blink detection, eye tracking, and vitals monitoring, the latest
learning called Convolutional networking is used for finding
start-up named as Ellcie Healthy has created a advanced smart
region of interest in the model. The model we used have
eyewear that integrates a drowsiness detection system.
several hidden layers which are present between single initial
Combination approaches use a multitude of sensors installed
and final layers used to take input and output. An operation
in vehicle, for example cameras, monitors of measuring heart
named Convolution is performed inside of these layer's using
beat rate, and infrared, in order to produce impressive results
matrix and several filters.
[14][15][16].
A deep learning technique which is used to take images and
In his paper, B. K. Rajan focuses on developing a system that
find some region of interests in it based on their distinctive
can identify driver weariness more precisely and accurately.
qualities and significance (learnable weights and biases)
Image processing is the foundation of this system. Image-
called Convolutional neural network/CNN. ConvNet needs a
based solutions are more secure and simple to use than
lot less pre-processing than other classification methods do.
vehicle-based and physiological signal-based techniques.
ConvNets are able to acquire these filters and attributes with
Measure of sleepiness of the vehicle's driver was identified by
enough practise, even though hand-engineered filters are
including mentioned method that depends on two major
utilised in simple implementations.
factors. In these factors, first test is measuring length taken by
the blink, and second one is counting the no. of blinks. Viola The architecture of convolutional model is made same like
jones and the KLT algorithm were combined to detect and structure of a human brain with the neurons connected
track the face [17]. together. This way of model formation usually gives an
average score but in case of images that contains heavy pixels
In his paper, J. Manikandan focuses on monitoring the mouth
it shows nearly zero accuracy.
and eyes to identify and categorize tiredness in a driver. For

643

Authorized licensed use limited to: AMITY University. Downloaded on June 06,2024 at 12:34:24 UTC from IEEE Xplore. Restrictions apply.
 2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N)

In order to develop the model to work more accurately we regularization, dimension reduction, factorized convolutions,
have used Inception V3, a technique used in deep learning and parallelized calculations [19][20].
using convolutional networks which uses a pre planned
training and then testing using some specific datasets which
helps in making the model more accurate.
Such a technique in which we take a large dataset and then
divide major part of it for training the model and the other part
for testing the accuracy of the trained model, and then using
such a model and using some filters and layers for detecting
specific type of object is called Transfer Learning. This idea
is that the convolutional layers are used to capture common,
low-level characteristics that apply to all images, such edges, FIGURE 2. Inception V3 Model
patterns, and curves, while the following layers identify
specific features inside an image, like eyes or wheels. Transfer IV. SIMULATION
learning is the process of using a model that has already been Proposed model has been simulated using Inception V3
learned to solve a new issue. Given that deep neural networks model which work on convolutional neural network
may be trained using little to no data at the moment, it is algorithm.
particularly well-liked in the field of deep learning. In this project we have used the dataset that is taken from
Media Research Lab(MRL) [21].
Using models developed for one problem as the foundation for
another requires this type of learning technique. Since it can Steps:
be used for any model or can be used with other models Step 1 – Take the Image as an Input with the help of Camera.
effortlessly. This can also be used for initial or earlier Webcam is using for taking different photograph and
processing and data gathering. Many of current best models enter photograph at different point of time. Therefore, an
on ImageNet's image recognition tasks, including VGG, infinite loop has been created by us, which take pictures and
Inception, and ResNet, are easily accessible using Keras. save every frame to access computer webcam. The OpenCV
CNN: The Convolutional Neural Network employs the is being employed using cv2 technique. To connect to smart
proposed method for detecting driver inattention (CNN). camera to the vehicle and configure the capture object (cap),
Comparatively speaking, CNN uses substantially less pre- use VideoCapture(0). cap.read() will be used for reading each
processing than other classification methods. Three different frame , and the image that has been taken will be saved in the
types of pooling, convolution, and fully linked layers often given frame variable.
make up the mathematical method known as CNN. Step 2 – Face will be found in given image and it developed
a region of interest (ROI) depend on face position in the
image.
As we know that the OpenCV algorithm is used for
identifying objects which are accepting grayscale images as
input, first we have to convert the image to grayscale image
to identify the face in it. Colour data is not required for
identifying the objects . Haar cascade classifier used in this
model to find faces. We set the classifier face =
cv2.CascadeClassifier(‘ path to our haar cascade xml file’).
After which the model is going to perform the detection by
using faces =face.detectMultiScale( gray , scaleFactor = 1.2 ,
minNeighbors = 3). It gives output a list of detections which
contain information the height, width, and x, y coordinates of
given object's boundary box. Now every faces can be iterated
over, and it can draw each face’s border boxes.
for ( x, y, w, h ) in faces:
FIGURE 1. CNN Architecture [Source: cv2.rectangle(frame,pt1=(x,y),pt2=(x+w,y+h),
https://www.upgrad.com/blog/basic-cnn-architecture/] color= (255,0,0), thickness=3 )

Inception V3: The Inception v3 model, that was released in Step 3 – Find eyes from developed ROI and pass it to our
2015, contains less errors than its predecessors and a total of classification model.
42 layers. Several network optimization techniques have been The method which is used for detecting faces is also used
suggested for an Inception v3 model to loosen the constraints for detecting eyes. The cascade classifier for the eyes is first
and simplify model adaptation. The methods include established then we detect the eyes by using eyes =

644

Authorized licensed use limited to: AMITY University. Downloaded on June 06,2024 at 12:34:24 UTC from IEEE Xplore. Restrictions apply.
 2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N)

eye_cascade.detectMultiScale(gray, scaleFactor = 1.1,

minNeighbors=1) . We will isolate the eyes' data from the
entire taken image. This can be done by getting the eye's
boundary box, due to which this code is using to extract the
eye's picture from the frame.
for (ex,ey,ew,eh) in eyes:
eye= frame[ey:ey+eh,ex:ex+w]
Step 4 – Categorization of classification model will depend
on whether eyes are open or closed. FIGURE 3. Activity Diagram of Driver Drowsiness
We are using Inception V3 which is CNN classification Detection System
model for predicting eye status. We carry out particular
operations for entering input our photographs into the model V. RESULT ANALYSIS
because for it the proper starting dimensions is necessary. As
After the model is trained using Inception V3 which is a CNN
the model we used was programmed on 80*80 pixel images,
model. We will get the performance table. Now, the best
we first resize the image to that size using cv2.resize(r_eye,
model (the one with the best testing accuracy and validation
(80,80)). This model normalize our input data eye = eye/255
accuracy) detects the drowsiness with:
(all of the values will range from 0 to 1) for better
84.27% accuracy on the testing sample set.
convergence.
95.09% accuracy on the training sample set.
eye= cv2.resize(eye,(80,80))
92.65% accuracy on the validation sample set.
eye= eye/255
These results are very good considering that the data is
eye= eye.reshape(80,80,3)
balanced.
eye= np.expand_dims(eye,axis=0)
Performance table of the model:
Enhance the dimensions to add to our classifier's input. We
started the developed model by using model = load_model
TABLE 1. Performance Table
(‘path\models\model.h5’). Now it will start predicting each
eye with the developed model.
Step 5 – The score has been calculated to determine whether
the person is sleepy or not.
The score that is used in model is basically a no. that will
be used to calculate and find for how much time the driver
has kept his eyes shut. Therefore, if both of the eyes are shut,
the score will keep rising, whereas an open eye causes the
score to fall. By using cv2.putText() function, the model is
drawing the outcome on the screen to show the real time
status of the person. FIGURE 4. Above graph showing loss in training sample
if(Score>15): dataset and in validation dataset against number of taken
try: epochs in model
sound.play()
except:
pass
Here we have established a threshold value in such a way that
whenever score value reaches beyond 15, then the system
shows that the person had their eyes closed for considerable
amount of time. And this is the point where alarm will start FIGURE 5. Above graph showing accuracy in training
playing its beep using sound.play function. sample dataset and in validation sample dataset against
Thus, using this method system finds whether the driver is in number of taken epochs in model
the drowsy state or not in this project.
VI. CONCLUSION AND FUTURE SCOPE
This project has developed by examining many existing
sleepiness detection algorithms from a number of different
perspectives. Although there have been a lot successful high
accuracy algorithms developed, there is still room for
advancement. Our project describes a better version of current
driver drowsiness detection system with the help of CNN. The
project has optimized the algorithm and used more filtered and
better dataset so that accuracy can be increased. Proposed

645

Authorized licensed use limited to: AMITY University. Downloaded on June 06,2024 at 12:34:24 UTC from IEEE Xplore. Restrictions apply.
 2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N)

project provides the training accuracy which is equal to [10] W. Deng and R. Wu, "Real-Time Driver-Drowsiness Detection
95.09% , the testing accuracy with the help to 84.27% and the System Using Facial Features," in IEEE Access, vol. 7, pp. 118727-
validation accuracy with the help to 92.65%. This project now, 118738, 2019, doi: 10.1109/ACCESS.2019.2936663.
saving time, and trying to reduce the reasons of accidents [11] A. D. McDonald, J. D. Lee, C. Schwarz, and T. L. Brown, “A
based on driver drowsiness by giving indication to the driver contextual and temporal algorithm for driver drowsiness detection,”
as an alert in form of alarm with better accuracy. In light of Accident Analysis & Prevention, vol. 113, pp. 25– 37, Apr. 2018.
the fact that this work uses a non-intrusive approach to detect [Online]. Available:
weariness from image sequences—a issue that is yet https://www.sciencedirect.com/science/article/pii/S0001457518300
058.
unsolved—the fundamental innovation of this work is its
ability to do so. [12] C. S. Wei, Y. T. Wang, C. T. Lin, and T. P. Jung, “Toward
Drowsiness Detection Using Non-hair-Bearing EEG-Based Brain-
REFERENCES Computer Interfaces,” IEEE Transactions on Neural Systems and
Rehabilitation Engineering, 2018.
[1] P. Wang and L. Shen, "A method of detecting driver drowsiness
state based on multi-features of face," 2012 5th International [13] V. J. Kartsch, S. Benatti, P. D. Schiavone, D. Rossi, and L.
Congress on Image and Signal Processing, Chongqing, China, 2012, Benini, “A sensor fusion approach for drowsiness detection in
pp. 1171-1175, doi: 10.1109/CISP.2012.6469987. wearable ultra-low-power systems,” Information Fusion, vol. 43, pp.
66–76, Sep. 2018. [Online]. Available:
[2] J. Yu, S. Park, S. Lee and M. Jeon, "Driver Drowsiness https://www.sciencedirect.com/science/article/
Detection Using Condition-Adaptive Representation Learning pii/S1566253517306942.
Framework," in IEEE Transactions on Intelligent Transportation
Systems, vol. 20, no. 11, pp. 4206-4218, Nov. 2019, doi: [14] M. Tayab Khan, H. Anwar, F. Ullah, A. Ur Rehman, R. Ullah,
10.1109/TITS.2018.2883823. A. Iqbal, B.-H. Lee, and K. S. Kwak, “Smart Real-Time Video
Surveillance Platform for Drowsiness Detection Based on Eyelid
[3] M. A. Zulkarnanie, K. S. Shanmugam, N. Badruddin and M. N. Closure,” Wireless Communications and Mobile Computing, vol.
M. Saad, "Enhancements to PERCLOS Algorithm for Determining 2019, pp. 1–9, 2019.
Eye Closures," 2022 International Conference on Future Trends in
Smart Communities (ICFTSC), Kuching, Sarawak, Malaysia, 2022, [15] L. Celona, L. Mammana, S. Bianco, and R. Schettini, “A multi-
pp. 76-81, doi: 10.1109/ICFTSC57269.2022.10039811. task CNN framework for driver face monitoring,” IEEE International
Conference on Consumer Electronics - Berlin, ICCE-Berlin, vol.
[4] Tian, Y., Cao, J. Fatigue driving detection based on 2018-Septe, pp. 1–4, 2018.
electrooculography: a review. J Image Video Proc. 2021, 33 (2021).
https://doi.org/10.1186/s13640-021-00575-1. [16] Y. Xie, K. Chen, and Y. L. Murphey, “Real-time and Robust
Driver Yawning Detection with Deep Neural Networks,”
[5] S. Yaacob, N. A. Izzati Affandi, P. Krishnan, A. Rasyadan, M. Proceedings of the 2018 IEEE Symposium Series on Computational
Yaakop and F. Mohamed, "Drowsiness detection using EEG and Intelligence, SSCI 2018, pp. 532–538, 2019.
ECG signals," 2020 IEEE 2nd International Conference on Artificial
Intelligence in Engineering and Technology (IICAIET), Kota [17] H. Joseph and B. K. Rajan, "Real Time Drowsiness Detection
Kinabalu, Malaysia, 2020, pp. 1-5, doi: using Viola Jones & KLT," 2020 International Conference on Smart
10.1109/IICAIET49801.2020.9257867. Electronics and Communication (ICOSEC), Trichy, India.

[6] W. Qing, S. BingXi, X. Bin and Z. Junjie, "A PERCLOS-Based [18] S. Mohanty, S. V. Hegde, S. Prasad and J. Manikandan,
Driver Fatigue Recognition Application for Smart Vehicle Space," "Design of Real-time Drowsiness Detection System using Dlib,"
2010 Third International Symposium on Information Processing, 2019 IEEE International WIE Conference on Electrical and
Qingdao, China, 2010, pp. 437-441, doi: 10.1109/ISIP.2010.116. Computer Engineering (WIECON-ECE), Bangalore, India, 2019.

[7] K. S. Sankaran, N. Vasudevan and V. Nagarajan, "Driver [19] Y. Zhao, K. Xie, Z. Zou and J. -B. He, "Intelligent Recognition
Drowsiness Detection using Percentage Eye Closure Method," 2020 of Fatigue and Sleepiness Based on InceptionV3-LSTM via Multi-
International Conference on Communication and Signal Processing Feature Fusion," in IEEE Access, vol. 8, pp. 144205-144217, 2020,
(ICCSP), Chennai, India, 2020, pp. 1422-1425, doi: doi: 10.1109/ACCESS.2020.3014508.
10.1109/ICCSP48568.2020.9182059. [20] A. Biju and A. Edison, "Drowsy Driver Detection Using Two
[8] A. Dasgupta, D. Rahman and A. Routray, "A Smartphone- Stage Convolutional Neural Networks," 2020 IEEE Recent
Based Drowsiness Detection and Warning System for Automotive Advances in Intelligent Computational Systems (RAICS),
Drivers," in IEEE Transactions on Intelligent Transportation Thiruvananthapuram, India, 2020, pp. 7-12, doi:
Systems, vol. 20, no. 11, pp. 4045-4054, Nov. 2019, doi: 10.1109/RAICS51191.2020.9332476.
10.1109/TITS.2018.2879609. [21] “MRL Eye Dataset | MRL,” MRL Eye Dataset | MRL.
[9] S. Kailasam, M. Karthiga, R. M. Priyadarshini, K. Kartheeban [Online]. Available: http://mrl.cs.vsb.cz/eyedataset.
and K. Anithadevi, "Accident Alert System for Driver Using Face
Recognition," 2019 IEEE International Conference on Intelligent
Techniques in Control, Optimization and Signal Processing
(INCOS), 2019, pp. 1- 4, doi: 10.1109/INCOS45849.2019.8951320.

646

Authorized licensed use limited to: AMITY University. Downloaded on June 06,2024 at 12:34:24 UTC from IEEE Xplore. Restrictions apply.