Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727

Wide Awake: Neural Network-Driven

Real-Time Drowsiness Detection
System for Enhancing Driver Safety
JPD Wijesekara1; HMJB Rathnayake2; Pavithra Subhashini3; Krishantha DKGK4
1
Department of Computer and Data Science, Faculty of Computing, NSBM Green University, Sri Lanka
2
Digital Engineering, Delivery, Virtusa Pvt Ltd, Sri Lanka
3
Department of Software Engineering, Faculty of Computing, NSBM Green University, Sri Lanka
4
Department of Innovations, Brandix Asia Holdings, Sri Lanka

Publication Date: 2025/02/19

Abstract: Drowsy driving is a critical issue contributing to a significant number of road accidents globally, resulting in
substantial loss of life and property. Traditional methods for detecting driver drowsiness, such as physical monitoring and
vehicle behavior analysis, have inherent limitations in terms of accuracy and practicality. This research focuses on
developing a drowsy driver detection system utilizing advanced neural networks, particularly convolutional neural networks
(CNNs), to analyze drivers' eye closure and behavior through real-time video input. The system's architecture comprises
several components, including a deep learning model trained on extensive image datasets, integrated with computer vision
and image processing technologies to enhance detection accuracy. Data collection involved diverse datasets of driver images
and videos under varying conditions to ensure robustness. The CNN model processes these images to classify the driver's
state of alertness. Experimental results demonstrated high accuracy, with precision and recall rates significantly
outperforming traditional methods. The system's ability to process real-time video input and accurately classify eye states
provides a robust solution for drowsy driver detection. The research discusses the methodology, training process,
implementation, and potential implications for improving road safety. Ethical considerations, such as ensuring driver
privacy and data security, are also addressed. Future work will focus on enhancing system robustness under various real-
world conditions and integrating additional data sources to improve detection accuracy. The Wide-Awake system represents
a promising advancement in leveraging deep learning and computer vision technologies to reduce the incidence of accidents
caused by driver fatigue.

Keywords: Drowsiness Detection, Driver Fatigue, Neural Networks, Convolutional Neural Networks (CNNs), Real-Time
Monitoring, Driver Safety, Computer Vision, Image Processing, Machine Learning, Facial Feature Detection, Eye State
Classification, Road Safety, Artificial Intelligence, OpenCV, Dlib, Deep Learning, Mobile Application, Web Application, Real-Time
Video Analysis, and Driver Monitoring System.

How to Cite: JPD Wijesekara; HMJB Rathnayake; Pavithra Subhashini; Krishantha DKGK (2025). Wide Awake: Neural Network-
Driven Real-Time Drowsiness Detection System for Enhancing Driver Safety. International Journal of Innovative
Science and Research Technology, 10 (1), 2698-2708. https://doi.org/10.5281/zenodo.14891727

I. INTRODUCTION Organization, approximately 1.35 million people die annually

due to road vehicle accidents, with a significant proportion
The project, titled Wide Awake, addresses the critical attributed to drowsy driving. The AAA Foundation for
issue of drowsy driving within the automotive industry. Traffic Safety estimates that 9.5% of accidents involve
Although there is no direct client, the project aims to develop fatigued drivers, highlighting the pervasive nature of this
an experimental solution that can be implemented in future issue. These incidents predominantly affect long-distance
applications to mitigate the risks associated with driver drivers, such as truck drivers, taxi drivers, and those traveling
fatigue. Drowsy driving poses a significant threat to road without adequate rest. Addressing this issue is critical to
safety, and current solutions have yet to achieve the desired improving road safety and reducing the economic impact of
accuracy and practicality necessary for widespread adoption. traffic accidents, which cost countries an estimated 3% of
their gross domestic product annually.
Drowsy driving is a leading cause of road accidents
worldwide, contributing to a substantial number of fatalities Drowsy driving accidents are often caused by drivers'
and severe injuries each year. According to the World Health diminished alertness, slower reaction times, and impaired

IJISRT25JAN1749 www.ijisrt.com 2698

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
decision-making abilities due to lack of sleep. Common neural networks (CNNs), computer vision, and image
scenarios include long-haul truckers driving through the processing, the Wide-Awake system aims to offer a more
night, shift workers commuting home after extended hours, accurate and practical solution than existing methods.
and families on long road trips without adequate breaks. The
consequences of such accidents are often devastating, leading The deep learning-based system leverages CNNs to
to loss of life, severe injuries, and significant property process video input from a camera mounted in the vehicle,
damage. detecting signs of drowsiness by analyzing eye closure rates
and patterns. The mobile application provides a convenient
solution for drivers who do not have the integrated system in
their vehicles, allowing them to use their smartphone cameras
for detection. The web application extends the system's
accessibility, enabling any device with a camera and internet
connection to utilize the drowsy driver detection capabilities.

The primary objective of the Wide-Awake project is to

enhance road safety by detecting and alerting drowsy drivers
using advanced technologies. The system aims to save lives
and prevent property damage by identifying drowsiness early
and prompting drivers to take necessary actions. The Wide-
Awake project seeks to address the limitations of current
drowsy driver detection methods by offering a solution that is
both accurate and practical for real-world implementation.

By focusing on the development and integration of deep

learning models, computer vision techniques, and user-
friendly applications, the project aims to create a
comprehensive drowsy driver detection system. This system
Fig 1 Percentage of Accidents cause of Drowsy Drivers. is designed to be adaptable to various vehicle types and driver
(Source: Jansen, 2018) behaviors, ensuring wide applicability and effectiveness. The
ultimate goal is to reduce the incidence of accidents caused
by driver fatigue, contributing to safer roads and
communities.

In summary, the Wide-Awake project represents a

significant advancement in the field of road safety, leveraging
cutting-edge technologies to address a critical issue. By
providing a robust and practical solution for drowsy driver
detection, the project aims to make a meaningful impact on
reducing traffic accidents and enhancing driver safety.

II. LITERATURE REVIEW

 Detection Strategy Focusing on Driver’s Performance

One approach to detecting drowsy drivers is by
analyzing their driving performance. Techniques such as lane
tracking and monitoring the distance between vehicles are
commonly used. These methods rely on sensors attached to
the steering wheel, gas pedal, and dashboard. Previous studies
Fig 2 Percentage of Property-Damaging Accidents cause of have shown that drowsy drivers exhibit specific patterns in
Drowsy Drivers. steering wheel movements and grip. For instance, Pilutti and
(Source: Jansen,2018) Ulsoy (1999) highlight that the variability in steering wheel
movements can be a reliable indicator of a driver’s state of
To tackle the problem of drowsy driving, the Wide- alertness. Furthermore, systems like those employed by major
Awake system has been developed. This system comprises automotive manufacturers, including Nissan and Mercedes-
three main components: a deep learning-based system, a Benz, use such performance-based indicators. However,
mobile application, and a web application. The primary these methods have inherent limitations. They are
function of these components is to detect drowsy drivers and significantly influenced by external factors such as road
alert them to prevent accidents. The system analyzes drivers' conditions and light intensity, and they may not detect
facial features, particularly eye closure and behavior, using drowsiness that has not yet affected vehicle performance
real-time video input to make decisions and issue warnings. (ARTAUD et al., n.d.).
By utilizing advanced technologies such as convolutional

IJISRT25JAN1749 www.ijisrt.com 2699

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
as a person becomes drowsy, while the power of the HF band
increases. Similarly, EEG signals can reveal drowsiness
through changes in the power spectrum of brain waves,
particularly the increase in alpha and theta bands and the
decrease in beta bands (Lin et al. 2005). However, these
methods require the driver to wear sensors or devices, which
is impractical for everyday use and can be intrusive.

Although these methods provide high accuracy in

controlled environments, their practical application in real-
world driving scenarios is limited due to the need for
continuous, real-time monitoring of physiological signals.
The inconvenience and discomfort associated with wearing
such monitoring devices make them less suitable for long-
Fig 3 Display while Nissan Driver Attention Alert helps term use by drivers.
combat drowsy driving with innovative system.
(Source: Nissan Driver Attention Alert helps combat drowsy  Computer Vision and Image Processing
driving with innovative system - Passport Nissan Blog, Computer vision and image processing techniques offer
2021) a more practical and non-intrusive solution. These methods
analyze facial features, particularly the eyes, to determine the
Additionally, technologies focusing on vehicle behavior driver’s level of alertness. Systems using infrared light to
can only detect drowsiness after it starts to impact driving detect eye reflections have shown promising results. For
performance. This delay can be critical, as early detection is example, García et al. (2010) developed a system that
essential for preventing accidents. Lavergne (1996) noted that categorizes driver drowsiness into four stages based on eye
these systems might not be effective in detecting drowsiness closure rate detected using infrared light. While effective at
on smooth, straight roads where lane departure or erratic night, these systems often struggle in daylight conditions.
steering behavior is less likely to occur.
Daylight-based systems utilize techniques like Viola-
 Detection Strategy Focusing on Driver’s Condition Jones and Haar cascades to detect facial features. The Driver
Detection strategies focusing on the driver's condition State Sensor (DSS) device by Seeing Machines is a
are divided into two main categories: physiological signals commercial product that uses these techniques to monitor
and computer vision. eyelid opening percentage (Seeing Machines 2021). Wu et al.
(2012) further advanced this by using AdaBoost classification
 Physiological Signals to detect faces and Support Vector Machines (SVM) to
Physiological signals involve monitoring identify eye states, implementing Local Binary Patterns
electrophysiological signals like EEG, heart rate variability (LBP) for enhanced accuracy. Figure 03 shows a flow chart
(HRV), pulse rate, and respiration rate. Chua et al. (2012) of the system.
explain that the LF/HF ratio in heart rate variability decreases

Fig 4 Flow Chart of the System

(Source : Wu et al. (2012)

Despite these advancements, current computer vision- importance of early detection of drowsiness to prevent
based systems face challenges such as variations in lighting accidents, which requires systems to be highly sensitive and
conditions, driver movements, and the need for continuous accurate.
camera calibration. Matsuo and Khiat (2012) emphasize the

IJISRT25JAN1749 www.ijisrt.com 2700

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
By leveraging neural networks and deep learning, the videos. These datasets included images of drivers in different
Wide-Awake system aims to overcome the limitations of states of alertness, under varying lighting conditions, and
existing methods. Neural networks, particularly with different facial characteristics to ensure robustness and
convolutional neural networks (CNNs), are well-suited for generalizability.
image classification tasks due to their ability to learn
hierarchical feature representations from raw pixel data The primary datasets utilized were from large-scale
(Goodfellow et al., n.d.). The Wide-Awake system utilizes repositories like the Yawning Detection Dataset and the
CNNs to analyze real-time video input from a camera Closed Eyes in the Wild (CEW) dataset. These datasets
mounted in the vehicle, detecting signs of drowsiness by provided a diverse range of images featuring drivers with
analyzing eye closure rates and patterns. various expressions, ethnicities, and in different
environments. This diversity is crucial for training a model
This approach offers several advantages. First, it that generalizes well to real-world scenarios.
eliminates the need for intrusive sensors, making it more
comfortable and practical for everyday use. Second, the Data preprocessing involved several steps to enhance
system can adapt to varying lighting conditions and driver the model's ability to generalize from the training data. The
behaviors, improving its robustness and reliability. Third, by system trains a neural network using a dataset of over 10,000
processing data in real-time, it can provide immediate alerts images depicting open and closed human eyes. This trained
to drowsy drivers, enhancing road safety. model is then utilized to detect signs of drowsiness in drivers.
The system processes real-time video footage captured by a
However, implementing a neural network-based system camera positioned in front of the driver. By leveraging the
also presents challenges. Training a neural network requires pre-trained neural network, it effectively identifies instances
a large and diverse dataset to ensure accuracy and of driver sleepiness. These steps included normalization,
generalizability. The system must be able to process real-time which scales pixel values to a standard range, and data
video input efficiently, necessitating powerful computational augmentation, which artificially expands the dataset.
resources. Additionally, privacy and data security concerns Augmentation techniques included rotations, flips, brightness
must be addressed, as the system involves continuous adjustments, and translations. These techniques simulate
monitoring of drivers. different real-world conditions such as changes in lighting,
driver head movements, and occlusions like sunglasses or
In conclusion, while traditional methods for detecting hats.
drowsy driving have limitations in accuracy and practicality,
advancements in computer vision and neural networks offer  Neural Network Architecture
promising solutions. The Wide-Awake system aims to The neural network model used in the Wide-Awake
leverage these technologies to provide a robust, non- system is a Convolutional Neural Network (CNN) designed
intrusive, and accurate method for detecting driver to classify the state of the driver’s eyes as open or closed.
drowsiness, addressing a critical need in road safety. CNNs are well-suited for image classification tasks due to
their ability to learn hierarchical feature representations from
III. METHODOLOGY raw pixel data.

 Data Collection The architecture comprises several convolutional layers

The effectiveness of the drowsy driver detection system followed by pooling layers, which extract and down sample
hinges on the quality and diversity of the dataset used for the features from the input images, respectively. The final
training the neural network. Data was collected from various layers are fully connected layers that output the classification
sources, including public datasets of driver images and result. The architecture is as follows.

Fig 5 Neural Network Architecture Diagram for Wide-Awake.

(Source: Author Compilation)

IJISRT25JAN1749 www.ijisrt.com 2701

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
 Input Layer: Processes the input image of the driver’s Accuracy measures the overall correctness of the model,
face. while precision and recall provide insights into the model’s
 Convolutional Layers: Three layers with 32, 32, and 64 performance in detecting positive instances of drowsiness.
filters, respectively with kernel size of 3, each followed The F1-score, the harmonic mean of precision and recall,
by a ReLU activation function. These layers extract offers a balanced evaluation metric, particularly useful in
spatial features from the images. imbalanced datasets where one class (e.g., non-drowsy) may
 Pooling Layers: Max-pooling layers to downsample the dominate.
feature maps, reducing the spatial dimensions and
computational load while preserving important features.  System Implementation
 Fully Connected Layers: Two dense layers with 128 and The system was implemented using Python, leveraging
2 neurons, respectively. The final layer uses a softmax libraries such as TensorFlow and Keras for neural network
activation function to output the probabilities of the eye development, and OpenCV for real-time video processing.
states (open or closed). After reviewing the information from Vision.fe.uni-lj.si
(n.d.), it was determined that the Eye Aspect Ratio (EAR) is
 Training and Validation an effective metric for detecting drowsy drivers using this
The neural network was trained using the Adam web component. The EAR is computed for both eyes in each
optimizer, which adjusts the learning rate during training to frame using a specific formula. When the EAR drops below
achieve faster convergence. The loss function used was 0.25, the system identifies the driver as potentially drowsy
categorical cross-entropy, suitable for the classification task. based on the decreased eye openness. This threshold allows
Training involved multiple epochs, where the model the system to reliably detect signs of sleepiness in drivers.
iteratively improved its parameters to minimize the loss The model was deployed on a local server for testing, with a
function. webcam capturing real-time video input of the driver’s face.
The system continuously analyzed the video feed, detecting
The dataset was split into training and validation sets, drowsiness based on eye closure patterns and issuing alerts
with 80% used for training and 20% for validation. This split through audio signals and visual indicators on the dashboard.
ensures that the model can generalize well to unseen data.
During training, techniques such as early stopping and The calculation takes into account six specific points on
dropout were employed to prevent overfitting. Early stopping the eye, labeled P1 through P6. These points are positioned at
halts training when the validation performance stops key locations around the eye to accurately measure the Eye
improving, while dropout randomly disables neurons during Aspect Ratio (EAR). The EAR is calculated using these
training to encourage the network to learn robust features. points to assess the openness of the eyes, which helps in
determining whether the driver is showing signs of
The model's performance was evaluated using metrics drowsiness.
such as accuracy, precision, recall, and the F1-score.

Fig 6 Pre-Trained Facial Landmark Detector in the Dlib Library

(Source: Rosebrock,2021)

IJISRT25JAN1749 www.ijisrt.com 2702

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727

Fig 7 Figure Illustrating Associated Calculations and Mapping to Identify EAR.

(Source: Vision.fe.uni-lj.si, n.d.)

 Data Flow

 Video Capture: A webcam or vehicle-mounted camera

captures real-time video of the driver's face.
 Frame Extraction: The video stream is divided into
individual frames, each representing a still image of the
driver.
 Face Detection: OpenCV is used to detect the driver’s
face within each frame. This step isolates the region of
interest (ROI) containing the driver’s eyes.
 Eye Detection: Within the ROI, the system detects the
driver’s eyes using pre-trained classifiers such as Haar
cascades.
 Eye State Classification: The detected eye images are
passed through the CNN, which classifies each eye as
open or closed.
 Drowsiness Detection: The system calculates the eye
closure rate over a sliding window of frames. If the eye
closure rate exceeds a certain threshold (e.g., eyes closed
for more than 30% of the frames in a 1-second interval),
the driver is classified as drowsy.
 Alert Generation: Upon detecting drowsiness, the
system triggers alerts. These alerts include audio signals
(e.g., a beep) and visual indicators (e.g., flashing lights on
the dashboard) to wake the driver.

OpenCV is employed to detect the facial features of the

driver, while Dlib, along with the SoundFile and Sounddevice
modules, is utilized to trigger the alarm.

 Tools and Technologies

 TensorFlow and Keras: Used for building, training, and

deploying the CNN model.
 OpenCV: Utilized for real-time video processing,
including face and eye detection.
 Flask: A micro web framework used to develop web
applications, enabling remote access to the detection
system.
 Android Studio: Used for developing mobile
applications, providing a platform for smartphone-based
detection.

Fig 8 Flow Chart of the Wide-Awake of the System.

(Source : Author Compilation)

IJISRT25JAN1749 www.ijisrt.com 2703

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
 System Components
The Wide-Awake system comprises three main
components: a deep learning-based system, a mobile
application, and a web application. Each component is
designed to detect drowsy drivers and alert them to prevent
accidents.

 Deep Learning-Based System: This component

leverages CNN to analyze real-time video input from a
vehicle-mounted camera. It provides the highest accuracy
due to its direct integration with the vehicle’s systems.
 Mobile Application: Designed for Android devices, this
app allows drivers to use their smartphone cameras for
drowsiness detection. It provides a convenient solution
for drivers who do not have the integrated system in their Fig 10 Screenshot taken from Web application.
vehicles. (Source: Author Compilation)
 Web Application: This component extends the system’s
accessibility, enabling any device with a camera and  Experimental Setup
internet connection to utilize the drowsy driver detection The experimental setup included followings
capabilities. The web application is built using Flask and
hosted on a cloud platform, ensuring scalability and ease  Participants: A diverse group of participants, including
of access. individuals of different ages, genders, and ethnic
backgrounds, to ensure the model's generalizability.
 Environment: Testing was conducted in both controlled
environments (e.g., indoor settings with consistent
lighting) and real-world driving scenarios (e.g., daytime
and nighttime driving).
 Equipment: The setup used high-definition webcams for
video capture and standard computers with sufficient
processing power to run the neural network in real-time.

IV. RESULTS

The experimental results demonstrated high accuracy in

detecting drowsiness, with significant improvements over
traditional methods. The key findings can be summarized as
follows, The system achieved an accuracy of 95%, indicating
that it correctly classified 95% of the instances. The precision
was 92%, meaning that 92% of the detections were true
positives. The recall rate was 90%, indicating that the system
successfully identified 90% of all drowsy instances. The F1-
score was 91%, reflecting a balanced performance in terms of
precision and recall.

These results highlight the effectiveness of the Wide-

Awake system in accurately detecting drowsiness, even under
varying conditions. The high precision and recall rates are
particularly important, as they indicate the system’s
reliability in both identifying drowsy drivers and minimizing
false alarms. In the neural network-based system depicted in
Figure 08, when the driver is alert, the Open Score should
range between 0 and 15. At this moment, the score is recorded
as 0, indicating that the driver's eyes are fully open, and they
are awake. Below is another screenshot from the neural
network-based system. Unlike the previous screenshot, this
one shows the driver asleep. Here, the Closed Score exceeds
15, with the current value being 19. This indicates that the
system has detected the driver is drowsy and has triggered the
Fig 9 Screenshot taken from Mobile Application.
alarm. Upon activation, a red border appears around the
(Source: Author Compilation)
window, as illustrated in the image below Figure 09.

IJISRT25JAN1749 www.ijisrt.com 2704

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
especially those produced by brands like Tesla, Mercedes-
Benz, and Volvo, already feature advanced driver assistance
systems (ADAS) that address various safety concerns. By
incorporating the Wide-Awake system, these vehicles can
achieve enhanced drowsiness detection capabilities that
complement and improve their existing safety technologies.

For traditional vehicles, which typically lack built-in

drowsiness detection systems, the Wide-Awake approach
provides an opportunity to retrofit safety features without the
need for expensive hardware installations. The system’s
reliance on standard cameras and its compatibility with
mobile and web platforms make it an accessible and cost-
effective solution. For example, drivers of older vehicle
models can use their smartphones as detection devices,
receiving real-time alerts and notifications without requiring
Fig 11 In the neural network-based system, when the driver any major modifications to their vehicles. This democratizes
is alert, the Open Score should range between 0 and 15. access to cutting-edge safety technology, ensuring that even
(Source: Author Compilation) vehicles without modern ADAS features can benefit from
enhanced road safety measures.

In the case of newer vehicles equipped with basic driver

monitoring systems, the integration of Wide-Awake can
significantly improve accuracy and early detection
capabilities. Existing systems often rely on predefined
parameters, such as steering wheel movements or heart rate
monitoring, which may not capture subtle signs of fatigue.
Wide-Awake’s use of convolutional neural networks (CNNs)
for real-time video analysis enables it to detect early
indicators of drowsiness, providing alerts before fatigue
translates into noticeable behavioral changes. This synergy
can enhance the overall safety profile of the vehicle, offering
a multi-layered approach to driver monitoring.

Integrating Wide-Awake into modern vehicles also

enables cross-platform functionality. For instance, data
Fig 12 This indicates that the system has detected the driver from Wide-Awake’s mobile or web applications can
is drowsy and has triggered the alarm. seamlessly integrate with a vehicle’s onboard systems,
(Source: Author Compilation) creating a unified safety ecosystem. Alerts from the system
could be displayed on the vehicle’s dashboard, alongside
 Comparison with Traditional Methods other driver assistance features, to provide a cohesive and
The system’s performance was compared with user-friendly experience. Such integration enhances driver
traditional methods such as physical monitoring and vehicle awareness and ensures that safety measures are consistently
behavior analysis. Traditional methods, including steering reinforced, regardless of the platform being used.
wheel movement monitoring and lane departure detection,
typically achieve lower accuracy rates due to their sensitivity One significant benefit of integrating Wide-Awake into
to external factors like road conditions and light intensity. traditional and modern vehicles is its ability to adapt to
diverse driving environments and user behaviors. While
For instance, traditional methods showed an average traditional systems may struggle with changing lighting
accuracy of around 75%, significantly lower than the 95% conditions, extreme weather, or driver variability, Wide-
accuracy achieved by the Wide-Awake system. Additionally, Awake’s advanced training techniques ensure reliable
traditional methods often struggle with early detection of performance across a wide range of scenarios. This
drowsiness, as they rely on noticeable changes in driving adaptability is particularly beneficial for long-distance
behavior, which may occur too late to prevent accidents. drivers, such as truck drivers or ride-share operators, who
often face extended hours on the road and varied driving
V. DISCUSSION conditions.

The Wide-Awake system not only surpasses existing Moreover, Wide-Awake’s non-intrusive design makes
drowsiness detection technologies in performance but also it an ideal addition to both traditional and modern vehicles.
offers significant benefits when integrated into both Unlike physiological signal-based systems that require
traditional and the latest vehicles. Modern vehicles, drivers to wear sensors, Wide-Awake operates seamlessly

IJISRT25JAN1749 www.ijisrt.com 2705

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
through video input, ensuring comfort and practicality. This  Ethical Considerations
design approach aligns with the convenience and user- Implementing a drowsy driver detection system raises
centricity expected by today’s drivers, particularly those several ethical considerations. Privacy is a primary concern,
accustomed to the intuitive interfaces of newer vehicles. as the system involves continuous monitoring of drivers. To
address this, the Wide-Awake system is designed to process
From a safety perspective, integrating Wide-Awake into video data in real-time without storing personal footage,
vehicle systems can reduce accident rates by ensuring real- ensuring that drivers' privacy is respected while maintaining
time detection and response. For example, the system can the system's functionality.
issue both audio and visual alerts to prompt the driver to take
corrective actions immediately upon detecting signs of Data security is another critical consideration. The
fatigue. In modern vehicles equipped with semi-autonomous system must safeguard the video data from unauthorized
driving capabilities, Wide-Awake could also work in tandem access and ensure that any transmitted data is encrypted.
with autonomous systems to temporarily take control of the Compliance with data protection regulations, such as the
vehicle or guide it to safety if the driver remains General Data Protection Regulation (GDPR), is essential to
unresponsive. protect user data and maintain trust.

In addition to improving safety, integrating Wide- Finally, it is important to consider the potential impact
Awake into traditional and modern vehicles contributes to of false positives and false negatives. False positives, where
their market appeal. As safety and advanced technology the system incorrectly identifies a non-drowsy driver as
become increasingly important differentiators for consumers, drowsy, could cause unnecessary anxiety or interruptions.
offering a cutting-edge drowsiness detection system like False negatives, where a drowsy driver is not detected, could
Wide-Awake can enhance a brand’s reputation and have severe consequences. The system’s design and testing
competitive edge. Automakers can market Wide-Awake as must strive to minimize these errors to ensure reliability and
part of their broader commitment to road safety and user acceptance.
innovation, appealing to safety-conscious consumers and
meeting stricter regulatory requirements in global markets. VI. CONCLUSION

In conclusion, the integration of the Wide-Awake The Wide-Awake project successfully demonstrates the
system into traditional and modern vehicles offers numerous potential of a neural network-based system for detecting
benefits. For older vehicles, it provides an affordable and drowsy drivers. By leveraging advanced technologies such as
accessible upgrade that enhances safety without requiring convolutional neural networks (CNNs), computer vision, and
significant hardware changes. For the latest vehicles, it real-time video analysis, the system provides a robust,
complements existing driver monitoring systems, delivering accurate, and non-intrusive method for identifying driver
higher accuracy and earlier detection capabilities. The fatigue. This approach addresses the critical issue of drowsy
system’s adaptability, non-intrusive design, and ability to driving, a major cause of road accidents worldwide, leading
integrate seamlessly with existing technologies make it a to significant loss of life and property. The system achieves a
transformative solution for reducing drowsy driving incidents high accuracy rate of 95%, significantly surpassing
and enhancing overall road safety. By bridging the gap traditional methods, with precision and recall rates of 92%
between cutting-edge technology and practical and 90%, respectively, ensuring reliable detection and
implementation, Wide-Awake sets the stage for a safer and minimizing the risk of false positives and negatives. The real-
more inclusive driving experience. time processing capability allows for immediate alerts,
crucial for preventing accidents, while its non-intrusive
 Challenges design enhances user comfort and practicality. The system's
Despite its advantages, the system also faces challenges adaptability to various lighting conditions and driver
that need to be addressed: behaviors further enhances its reliability. However,
challenges such as variability in driver behavior, extreme
 Variability in Driver Behavior: Individual differences lighting conditions, and computational efficiency need to be
in driver behavior and facial features can affect the addressed. Ethical considerations, including privacy and data
system’s performance. Continuous learning and security, are also crucial, with the system ensuring real-time
adaptation are necessary to handle such variability. processing without storing personal footage and complying
 Lighting Conditions: Although the system performs well with data protection regulations. The Wide-Awake system
under various lighting conditions, extreme variations represents a significant advancement in road safety, with the
(e.g., very low light or intense glare) can impact accuracy. potential to save lives and prevent property damage. Future
Further enhancements in image processing techniques are research will focus on enhancing robustness, integrating
needed to mitigate these effects. additional data sources, and optimizing performance. The
 Computational Requirements: Real-time video system's accurate, reliable, and user-friendly solution has the
processing requires significant computational resources. potential to reduce the incidence of accidents caused by driver
Optimization of the neural network model and efficient fatigue, making roads safer worldwide
use of hardware resources are essential for practical
deployment.

IJISRT25JAN1749 www.ijisrt.com 2706

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
REFERENCES [11]. "CNN Based Face Detector from Dlib." Medium.
Accessed June 24, 2024.
[1]. Distefano, Salvatore, Nilanjan Banerjee, and Antonio https://towardsdatascience.com/cnn-based-face-
Puliafito. "Smart Objects, Infrastructures, and detector-from-dlib-c3696195e01c.
Services in the Internet of Things." International [12]. "Detect Facial Features in Photos | Mobile Vision |
Journal of Distributed Sensor Networks 2016 (2016): Google Developers." Google Developers, 2021.
1-2. Accessed June 24, 2024.
[2]. Firdhous, Mohamed F. M., B. H. Sudantha, and P. M. https://developers.google.com/vision/android/detect-
Karunaratne. "IoT-Powered Sustainable Dry Zone faces-tutorial.
Agriculture: An Experimental Implementation." In [13]. "Driver Drowsiness Detection." Bosch Mobility
Proceedings of the 2018 International Conference on Solutions, 2021. Accessed June 24, 2024.
Information Technology Research (ICITR), 1-6. https://www.bosch-mobility-
IEEE, 2018. solutions.com/en/products-and-services/passenger-
[3]. Mangiaracina, Roberto, Giovanni Miragliotta, cars-and-light-commercial-vehicles/driver-
Alessandro Perego, Giulia Salvadori, and Alessandro assistance-systems/driver-drowsiness-detection/.
Tumino. "Adopting the Internet of Things [14]. "Driver Monitoring Technology for Automotive,
Technologies in Agriculture: An Analysis of the Transport & Logistics Industries." Seeing Machines,
Benefits in the Organic Wine Industry." In XXI 2021. Accessed June 24, 2024.
Summer School "Francesco Turco" - Industrial https://www.seeingmachines.com/.
Systems Engineering, 48-53. Naples, Italy, 2016. [15]. García, I., S. Bronte, L. Bergasa, N. Hernandez, B.
[4]. Qin, Pei, Zhiwei Lu, and Tao Zhu. "Application Delgado, and M. Sevillano. "Vision-Based
Research on Agricultural Production Throughout the Drowsiness Detector for a Realistic Driving
Internet of Things." In 3rd International Conference Simulator." 13th International IEEE Conference on
on Management, Education, Information and Control, Intelligent Transportation Systems, 2010, 887-894.
1219-1224. Shenyang, Liaoning, China, 2015. doi:10.1109/ITSC.2010.5625097.
[5]. Zhu, Xiang, and Yifan Lin. "ZigBee Implementation [16]. Jansen, Bart. "Drowsy Driving Plays Larger Role in
in Intelligent Agriculture Based on Internet of Accidents Than Federal Statistics Suggest." USA
Things." In 2nd International Conference on Today, February 8, 2018. Accessed June 24, 2024.
Electronic & Mechanical Engineering and https://www.usatoday.com/story/news/2018/02/07/aa
Information Technology, 1842-1846. Shenyang, a-drowsy-driving-plays-larger-role-accidents-than-
Liaoning, China, 2012. federal-statistics-suggest/313226002/.
[6]. "Over 45,000 Hectares of Paddy Cultivated in [17]. Lavergne, C. D. "Results of the Feasibility Study of a
Anuradhapura District During Yala Season." Daily System for Warning of Drowsiness at the Steering
News. May 25, 2022. Accessed June 24, 2024. Wheel Based on Analysis of Driver Eyelid
http://archives1.dailynews.lk/2022/05/25/local/45k- Movements." Proceedings of the Fifteenth
hectares-paddy-cultivated-anuradhapura. International Technical Conference on the Enhanced
[7]. Jayasundara, D. M., M. G. L. C. K. Weerakoon, and Safety of Vehicles, Held Melbourne, Australia, 13-16
D. M. L. D. Herath. "An Evaluation of Rice May 1996.
Production Trends in Sri Lanka: A Review." Sri Lanka [18]. Lin, Cheng-Ju, Chih-Hsiang Lin, and Shang-Hua
Journal of Food and Agriculture 4, no. 1 (2018): 1-14. Wang. "Integrated Image Sensor and Light
Accessed June 24, 2024. Convolutional Neural Network for Image
http://www.slcarp.lk/reports/production-trends. Classification." Mathematical Problems in
[8]. Artaud, P., S. Planque, C. Lavergne, H. Cara, P. de Engineering, 2021. doi:10.1155/2021/5573031.
Lepine, C. Tarriere, and National Highway Traffic [19]. Lin, Chin-Teng, Tzyy-Ping Jung, Ruey-Song Wu,
Safety Administration. "An On-Board System for Sheng-Fu Liang, and Tsorng-Yan Huang. "Estimating
Detecting Lapses of Alertness in Car Driving." Driving Performance Based on EEG Spectrum
Proceedings of the Fourteenth International Technical Analysis." EURASIP Journal on Advances in Signal
Conference on Enhanced Safety of Vehicles (94-S2- Processing 2005, no. 19 (2005).
O-08). doi:10.1155/asp.2005.3165.
[9]. "ATTENTION ASSIST: Drowsiness-Detection [20]. Matsuo, Hiroshi, and Ali Khiat. "Prediction of Drowsy
System Warns Drivers to Prevent Them Falling Driving by Monitoring Driver’s Behavior." 21st
Asleep Momentarily." Media Daimler, 2021. International Conference on Pattern Recognition
Accessed June 24, 2024. (ICPR 2012), 2012.
https://media.daimler.com/marsMediaSite/ko/en/936 [21]. Pilutti, Thomas, and Ali Ulsoy. "Identification of
1586. Driver State for Lane-Keeping Tasks." IEEE
[10]. Chua, E.-P., W.-Q. Tan, S.-C. Yeo, P. Lau, I. Lee, I. Transactions on Systems, Man, and Cybernetics - Part
Mien, and J. Gooley. "Heart Rate Variability Can Be A: Systems and Humans 29, no. 5 (1999): 486-502.
Used to Estimate Sleepiness-Related Decrements in doi:10.1109/3468.784175.
Psychomotor Vigilance During Total Sleep [22]. "Rapid Object Detection Using a Boosted Cascade of
Deprivation." Sleep, 2012. doi:10.5665/sleep.1688. Simple Features." IEEE Xplore, 2021. Accessed June

IJISRT25JAN1749 www.ijisrt.com 2707

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14891727
24, 2024.
https://ieeexplore.ieee.org/document/990517.
[23]. "Road Traffic Injuries." World Health Organization,
2021. Accessed June 24, 2024.
https://www.who.int/news-room/fact-
sheets/detail/road-traffic-injuries.
[24]. Rosebrock, Adrian. "Facial Landmarks with Dlib,
OpenCV, and Python." PyImageSearch, 2021.
Accessed June 24, 2024.
https://www.pyimagesearch.com/2017/04/03/facial-
landmarks-dlib-opencv-python/.

IJISRT25JAN1749 www.ijisrt.com 2708