Drivers Drowsiness Detection System
Richa Vivek Gupta*
Kadugula Priscilla Ritika Prakash
Sharda University,Uttar Pradesh
Sharda University,Uttar Pradesh Sharda University,Uttar Pradesh
Abstract— Accidents involving motor vehicles and other *communicating author
connected incidents frequently result in fatalities and serious
injuries among people. According to World Health
The Bosch-supplied driver fatigue detection
Organization estimates from the last five years, fatal crashes system makes choices using information from
caused by traffic-related injuries resulted in over 1.25 million the steering-wheel-mounted sensor, the vehicle's
deaths worldwide or about one crash every 25 seconds. One of speed, the use of its turn signals, and the lane
the major contributors to traffic accidents is drowsy driving. assist camera located in the front of the vehicle.
Road accidents can hence be decreased by identifying drowsy
drivers. A machine learning method for drowsiness detection is
described in this research. Recently, Tata Motors2, India’s largest commercial
vehicle manufacturer launched India's first CNG-powered
The areas of the driver's eyes are located using face Medium & Heavy Commercial Vehicle (M&HCV) truck. It
detection, and these regions serve as templates for eye tracking introduced safest trucks with Advanced Driver Assistance
in following frames. Finally, the detection is performed on the System (ADAS) – Collision Mitigation System, Lane
tracked eye pictures to produce alarm warnings. This method's Departure Warning System(LDWS) and Driver Monitoring
three stages are detection of Face, detection of Eyes, and
System.
detection of Drowsiness. The driver's face is recognized using
image processing, which then extracts the image of the driver's
eyes to check for signs of fatigue.
It should be noted that while these drowsiness detection
Keywords—Drowsiness Detection, Machine Learning, Image safety systems are typically found in high-end vehicles, their
Processing use is limited and not very common among drivers. Surveys
conducted in 20153 revealed a growth in the embedding and
I. INTRODUCTION connection of smart devices with sensors and mobile
operating systems like Android, which is the operating
Sleep deprivation contributes significantly to vehicle system with the largest installed base in automobiles.
accidents and has a similar impact on the brain as drinking. Additionally, deep learning has made ground-breaking
Driving while in this condition results in slow, decreased strides in machine learning during the past few years. As a
reaction time and stimulus judgment which leads to the result, implementing these new technologies and approaches
increased number of accidents. can be a useful way to both boost the effectiveness of the
Facts1 about fatigued driving: real-time driver sleepiness detection system that is already in
place and give drivers a tool that they can use on a regular
1. According to a 2005 survey by the National basis.
Sleep Foundation (NSF), 38 percent of adults
have dozed off behind the wheel and about 21 A machine learning method for drowsiness detection is
percent have led to fatal accidents. described in this research.
2. The majority of sleep-related accidents occur in Machine learning is an application of AI
young adults (18 to 29). (Artificial Intelligence) which uses algorithms to
separate data, learn from that data, and then put
3. Each year, car accidents result in approximately together rational decisions.
1,550 fatalities, 71,000 injuries, and about 1250
crores of financial loss. DL(Deep learning) is subfield of machine
learning that structures algorithms in layers to
The creation of drowsiness detection technology is a form an "artificial neural network" that can learn
challenge for both industry and academia. on its own and make intelligent decisions.
Volvo created the Driver Alert Control for the However, its capabilities are different. While
automotive sector, which uses a camera basic machine learning models get progressively
mounted on the car linked to its lane departure better at their task, they still require some
warning system to alert drivers who may be guidance. If an AI algorithm gives a wrong
driving while fatigued (LDWS). prediction, an engineer will have to step in and
make adjustments. With a deep learning model,
In a similar line, Mercedes-Benz has created and an algorithm can determine on its own if a
introduced the Attention Assist System, which prediction is accurate or not through its own
continuously assesses whether the information neural network. A deep learning model is
gathered from a driver's driving behavior designed to continually analyze data with a logic
matches with the steering movement and the structure similar to how a human would draw
current driving situation. conclusions.
Real-Time Shruti Pandey Haar 80%
Driver- Rushikesh cascade
Drowsiness Nikam 2014 classifier,
4 Detection Open CV
II. PROPOSED SYSTEM System
Using Facial
The fatigue detection system's goal is to help reduce Features
accidents involving both passenger and commercial vehicles. Driver Mahek Jain, eSpeak module, 82.4%
Before the driver completely loses all attention, the system drowsiness Bhavya Eye aspect ratio,
will recognize the early signs of tiredness and alert them that monitoring Bhagerathi, CNN
5
they can no longer operate the car safely. However, using based on Sowmyarani
this device won't ensure that the driver is fully aware and that yawning C N 2015
an accident won't occur. It is merely a tool for enhancing detection
driver safety, with a particular emphasis on long-distance Driver Wanghua Convolutional 90%
truck drivers, nighttime drivers, solo drivers, and those who Drowsines Deng, neural
are sleep deprived. s Detection Ruoxue Wu network,
by 2017 fatigue
Applying detection,
Deep feature
III. LITERATURE REVIEW 6 Learning location
Several approaches have been suggested in an effort to Techniques
improve drowsiness detection, accuracy and speed. The to
existing fatigue monitoring techniques and strategies are Sequences
summarized in this section. The first method that has been of
utilized is based on driving patterns, and it is very dependent Images
on the characteristics of the vehicle, the state of the roads, A Driver Shabnam Drowsiness 83%
and driving abilities. Drowsiness Abtahi , Monitoring,
Detection Behnoosh Hariri, face tracking,
Deviation from a lateral or lane position or the movement 7 Framework Shervin yawning
of the steering wheel should be taken into consideration Using Deep Shirmohammadi detection
when calculating the driving pattern4. The benefit of adopting Learning 2017
steering behavior for fatigue detection is that these devices Heuristic
monitor continuously, affordably, discreetly, and robustly A Uche M. Open CV, 80%
even under the most challenging environmental conditions. Drowsiness Chikezie 2017 Raspberry pi,
detection IoT
8
system using
deep learning
TABLE I. LITERATURE REVIEW TBLE and IoT
An Jongseong Hybrid 82.07%
S.No Accurac
Paper Title Author Method Used Investigation Gwak Akinari Sensing,Python
. y
of Early Hirao 2018 Open CV
Driver D.Mounika CNN,Android 82% Detection of
Drowsiness K. Divya Deepika studio, python Driver
Detection 2011 Drowsiness
Model Using 9 Using
Convolutional Ensemble
1 Neural Machine
Networks Learning
Techniques Based on
for Android Hybrid
Application Sensing
Driver M. Kesava Machine learning, 87% IoT-Based Keyong Li IoT 81%
Drowsiness Puneeth Varma Artificial Smart Alert Lisheng Jin Tensorflo
Detection 2012 Intelligence, System for 2018 w OpenCV
2 System Using Python 10
Drowsy
Computer Driver
Vision Detection
Real-Time Drashti Patel, CNN using 92% Effects of Yuying CNN 95%
Driver Rishika Tiwari tensorflow driver JiangLin Lin Deep Learning
Drowsiness 2012 behavior style Gao Machine
3 Detection differences 2018 learning
using 11 and individual
Computer differences on
Vision driver
sleepiness
detection
Drowsiness Anil Kumar Keras software 96% C. Proposed Design
monitoring Biswal Open CNN The physiological properties and the gradual change
based on Debabrata
12
driver and in the facial features like6- fluctuations in the driver's
Singh 2019
driving data concentration, changes in the size or position of the
fusion pupils and the eyelids, and rapid shifts in head
Driver Binod Tensorflo 87.6% position, head orientation, and blink rate are some of
drowsiness Kumar w Open the factors which are considered while detecting the
monitoring Pattanaya CV alertness level of the driver.
system using k
13
visual The device will be able to distinguish between tired
Ming-Hour
behavior and and awake drivers by analyzing the eye states. The
Yang 2020
machine driver's faces will continuously provide a video
learning stream, which will be fed into a microcontroller for
Drowsiness Motoki Shino Python 79% processing. After then, classifiers will be applied to
monitoring 2020 Deep learning categorize the condition of the driver's eye. If the
based on driver is found to be drowsy, a warning will be raised
14
driver and in the form of an alarm.
driving
behaviour Lane tracking detection algorithm:Image processing
Driver Hong J. EohMin Open CV 85% techniques are used to measure how a driven vehicle
Inattention K. Face behaves in relation to its position in relation to its
Monitoring ChungSeong- tracking surroundings.
15 System for Han Kim
Intelligent
2020
Vehicles: A
Review
IV. METHODOLOGY
The Drowsiness Detection System was created via an
ongoing cycle of study and analysis5. Concepts are created
during the research stage and requirements ,constraints are
examined during the analysis stage. The cycle is then
repeated to produce increasingly complex theories, which are
later examined in more detail.
A. Requirements
In order to fulfill its role as a system for increasing
driver safety, the solution must be able to identify
drowsiness with accuracy.
It's crucial that the solution is as transparent as
possible to the driver.
There are currently existing solutions to this issue, but
the best ones are typically too expensive to be widely
used. So, the solution must be economical.
The system should give real time response in order to
tackle high speed vehicles.
It should be flexible i.e. it should accommodate all
types of users.
B. Constraints Fig.1 Component Diagram
(Drivers drowsiness detection system)
The solution must be built to function properly with
limited power requirements.
The solution must be implemented in a space-
efficient way. It must not compromise the vehicle's
existing controls.
V. DATASET VI. WORKFLOW
We used the Real-Life Drowsiness Dataset to identify
multi-stage drowsiness for our training and test data. The A webcam is designed to track the driver's real-time,
ultimate goal is to make it possible for our system to live video which is subsequently edited into video
distinguish between milder forms of fatigue and more severe, frames.
evident occurrences of sleepiness.
Sometimes because of the constant alterations in the
About 2900 files are taken both of open and closed eyes amount of light, each frame of the video is
along with the rest of the facial features like preprocessed to meet the requirements (like
yawning.Many subjects were recorded in a range of illumination enhancement).
simulated driving scenarios, including standard driving
mode, yawning, slow blink rate, conscious laughter, and [For the project images/videos are extracted from the
dizzy napping, etc., during both day and night light settings. provided dataset].
As a result, we were able to compile enough data to take After that, Open CV(Open Source Computer Vision)
both alertness and tiredness into account. analyses the footage and finds any faces.
The DLIB is used to reanalyze the identified faces to
identify facial landmarks using the 68 point
technique. The key feature sets are extracted as Eyes
and Mouth(DLIB offers a pre-defined facial
landmark detector which is capable of detecting 68
point on the face).
Fig.2 Type of data used
(Dataset Distribution7)
Fig.4 68 point view8
From the 68 points, the coordinates of left eye, right
eye and mouth are located and are further provided
to the conditional logic for monitoring of EAR and
MAR values.
When the EAR value falls below the threshold value
for the predetermined amount of time, an alarm
sound is produced to warn the driver of their
drowsiness.
Fig.3 Dataset
VII. RESULT Improvements: Environmental lighting can be handled in
In this study, five driving simulation scenarios (with order to enhance this technology by adding a module that
glasses, with sunglasses, without glasses, at night with can measure the illumination levels and alter the threshold
glasses, and at night without glasses) were taken into value for blink detection correspondingly. A transmission
account. For each scenario, videos of two different states— module can also be integrated to convey the driving status
sleepy and awake—were chosen. With the help of dlib information in real-time to the appropriate authorities,
framework facial landmarks were created which intern were improving control and increasing driver situation
used for the fatigue detection. monitoring.
The results for various categories are provided below:-
ACKNOWLEDGMENT
The key elements concentration, dedication, hard work
and application are not only essential factors for achieving
the desired goals but also guidance, assistance and co-
operation of people is necessary.
We are grateful to our faculty guide Prof. Shaveta Khepra
for guiding us. She has the qualities to be an ideal guide. She
will always prevail upon our remembrance.
We are very fortunate to have unconditional support from
TABLE II. Accuracy in different cases our family. We thank our parents, who gave us the courage to
get our education, supported us in all achievements
throughout our life. Without their encouragement, this work
would indeed have been very difficult for us to tackle.
According to the evaluation outcomes of various
scenarios, eyes are essential for classifying tiredness in any
REFERENCES
situation. Additionally, it has been demonstrated that the
model's effectiveness decreases when the driver is wearing
sunglasses since the algorithm cannot detect the driver's eyes. [1] Drowsy Driving NHTSA reports. (2018, January 08). Retrieved
from https://www.medindia.net/patientinfo/drowsy-
driving.htm#:~:text=21%20percent%20of%20all%20fatal,drowsy
%20in%20the%20past%20year
When the system is tested under different lighting [2] Information on tata motors taken from
circumstances, the results continuously change based on the https://www.tatamotors.com/press/tata-motors-makes-indias-trucks-
amount of light. Lighting is a factor that has a significant smarter-safer-and-more-efficient/.
impact on the system's performance.
[3] T Cornez, R Cornez
Android Programming Concepts., Jones & Bartlett Publishers (2015)
[4] Krajewski, J. & Sommer, D. & Trutschel, U. & Edwards, D. & Golz,
VIII. CONCLUSION AND FUTURE SCOPE M., (2009) “Steering Wheel Behavior Based Estimation of
Fatigue”, Driving Assessment Conference 5(2009), 118-124.
The enormous potential of image processing is doi: https://doi.org/10.17077/drivingassessment.1311
demonstrated in this study. The majority of global problems
can be solved by using technology that is competent and [5] https://uwaterloo.ca/systems-design-engineering/current-
inexpensive for manipulating images, much to how humans undergraduate-students/courses/workshop-projects/fourth2002/
use vision to interact with the world. This will enable drowsiness-detection-system
machines to find solutions to various issues. [6] http://www.ijaema.com/gallery/241-ijaema-april-3854.pdf
The system's job is to identify facial landmarks in photos [7] https://www.kaggle.com/datasets/dheerajperumandla/
and send the collected information to the trained model to drowsiness-dataset?select=train
[8] https://www.irjmets.com/uploadedfiles/paper/
determine the driver's state. Given that present applications issue_4_april_2022/21084/final/fin_irjmets1650438819.pdf
cannot be used in embedded systems due to their restricted
computation and storage capacity, the method's goal is to
reduce the model's size. .
