Beyond Human Vision: AI‑Powered Eye Tracking for Safety and
Performance System
B. Venkata Charan Kumar, B. Megha Shyam Kumar, Y. Subbarayudu, S. Sandeep,
U. Prashanth and M. Mahindra Reddy
Department of Computer Science and Engineering (Data Science), Santhiram Engineering College,
Nandyal 518501, Andhra Pradesh, India
Keywords: Blink Rate Analysis, Facial Expression Recognition, Vigilance Monitoring, Human Factors, Physiological
Measurement, Computer‑Aided Diagnosis, Attentional State Monitoring, Multimodal Biometrics, Blink
Pattern Analysis, Human Activity Recognition, Behavioural Biometrics, Gaze Tracking, Eye Movement
Analysis, Facial Geometry, Medical Diagnostics.
Abstract: This study presents a novel real-time approach to detecting eye blinks by utilizing computer vision techniques
and geometric analysis of facial landmarks. The system employs OpenCV and Dib libraries to process video
input, recognize facial features, and accurately identify instances of eye closure. By analyzing the ratio
between the vertical and horizontal distances of specific eye landmarks, we develop a robust metric for blink
detection that remains effective across diverse lighting conditions and facial angles. Experimental testing
demonstrates a detection accuracy of 94.3% at 27 frames per second on standard hardware, making this
solution viable for practical applications. The implemented system shows particular promise in driver fatigue
monitoring systems, assistive technology interfaces, and clinical assessment of blinking patterns. This work
contributes to the growing field of non-intrusive behavioral monitoring by providing an efficient, accessible
method for eye blink detection that balances computational demands with real-time performance
requirements.
1 INTRODUCTION
1.1 Background and Motivation
The proliferation of intelligent vision-based systems
has paved the way for innovative applications in
human health monitoring and safety. Among these,
eye blink detection has emerged as a critical tool in
areas such as drowsiness detection, medical
diagnostics, and human-computer interaction.
Research indicates that abnormal blinking patterns
can be indicative of neurological disorders, fatigue, or
cognitive load, making automated blink analysis a
vital area of study. Recent advancements in facial
landmark detection, particularly through Dib’s pre-
trained models, have enabled real-time eye blink
monitoring with high accuracy. Furthermore, the
integration of computer vision with machine learning
enhances the reliability of detecting drowsiness in
drivers, assessing fatigue levels in workers, and
supporting medical diagnostics.
1.2 Problem Statement
Despite advancements in vision-based monitoring
systems, several challenges persist in real-time eye
blink detection:
• Environmental Variability: Changes in
lighting conditions and occlusions (e.g.,
glasses, hair) impact detection accuracy.
• Drowsiness Detection Robustness:
Traditional methods rely on subjective reports
or heuristic rules, leading to inconsistencies.
• Medical Application Viability: Most
existing systems are designed for general
fatigue detection and lack adaptability for
medical conditions like dry eye syndrome or
neurological disorders.
Our framework addresses these issues through:
1. A Dib-based eye landmark detection model
for precise real-time blink extraction.
2. A threshold-based and machine learning
Kumar, B. V. C., Kumar, B. M. S., Subbarayudu, Y., Sandeep, S., Prashanth, U. and Reddy, M. M.
Beyond Human Vision: AI-Powered Eye Tracking for Safety and Performance System.
DOI: 10.5220/0013886700004919
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 2, pages
579-587
ISBN: 978-989-758-777-1
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.
579

hybrid approach to detect drowsiness
accurately.
3. An adaptive preprocessing pipeline to
enhance robustness against environmental
variations.
1.3 Objectives of the Study
This research aims to achieve the following key
objectives:
• To develop a real-time eye blink detection
system with minimal computational overhead.
• To implement an efficient blink count extraction
algorithm for analyzing blinking patterns.
• To optimize the model for drowsiness
detection with potential applications in driver
safety and medical diagnostics.
• To validate the system’s effectiveness through
comparative studies with existing fatigue
detection techniques.
1.4 Contribution of the Study
• Novel Framework: The first implementation
of a Dib-based adaptive blink detection system
optimized for multiple applications.
• Enhanced Drowsiness Analysis: Incorporation
of dynamic blink frequency assessment to detect
fatigue more effectively.
• Computational Efficiency: Lightweight
implementation suitable for embedded and real-
time processing applications.
• Medical and Safety Applications: Potential
use cases in driver safety systems, neurological
disorder diagnosis, and ophthalmology
research.
This study bridges the gap between real-time eye
blink detection and its diverse applications, ensuring
a more reliable and efficient approach for drowsiness
monitoring and medical diagnostics. Figure 1 shows
the Eye Blink Detection Workflow.
Figure 1: Eye blink detection workflow.
2 LITERATURE REVIEW
2.1 The Evolution of Eye Blink
Detection Technology
The journey of eye blink detection technology has
evolved significantly over the past decades. In its
early stages (2000-2010), eye tracking systems relied
on infrared sensors and heuristic algorithms, capable
only of detecting exaggerated blinking patterns.
These rudimentary methods were limited by their
reliance on specific hardware setups and controlled
environments. The introduction of machine learning-
based approaches (2010-2018) marked a
breakthrough, allowing for improved generalization
across different facial structures and lighting
conditions. More recently, advancements in deep
learning and landmark-based techniques, such as
ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,
COMMUNICATION, AND COMPUTING TECHNOLOGIES
580

those implemented using Dib, have enabled real-time,
high-accuracy blink detection in natural settings,
making the technology more accessible for
applications in drowsiness detection, medical
diagnostics, and assistive technologies.
2.2 Eye Blink Detection in Fatigue and
Medical Diagnosis
Eye blink analysis has been widely used in fatigue
detection, particularly for monitoring driver
drowsiness. Studies have shown that prolonged eye
closure and irregular blinking patterns are strong
indicators of reduced alertness. Traditional fatigue
detection systems relied on vehicle-based sensors,
while modern approaches leverage facial landmark
tracking to achieve higher precision.
Beyond fatigue detection, eye blink metrics have also
been utilized in medical diagnostics, aiding in
conditions such as dry eye syndrome, Parkinson’s
disease, and neurological disorders. By integrating
blink frequency analysis with machine learning,
researchers have developed automated screening
tools capable of detecting early symptoms of these
conditions.
Figure 2 shows the Eye Difference ratio.
With continued advancements in AI, eye blink
detection systems are poised to offer even more
precise and personalized assessments for both fatigue
monitoring and medical diagnostics.
In clinical settings, eye blink analysis is becoming a
non-invasive tool to monitor patients with
neurological impairments, such as those recovering
from strokes or managing degenerative diseases.
Figure 2: Eye difference ratio.
2.3 Challenges in Real-Time Blink
Detection
Despite its advancements, real-time eye blink
detection faces several challenges. Lighting
variations can significantly impact the accuracy of
facial landmark detection, as poor illumination may
obscure key facial features. Obstructions such as
eyeglasses or facial hair pose additional difficulties in
accurately tracking eye movement. Interpersonal
variability, including differences in blink rates and
facial structures, necessitates adaptable models that
can generalize across diverse populations.
Addressing these challenges requires robust
preprocessing techniques, adaptive algorithms, and
continuous improvements in computer vision
methodologies to ensure accurate and reliable blink
detection across real-world scenarios.
3 METHODOLOGY
Our eye blink detection system leverages advanced
computer vision techniques and geometric analysis to
provide a robust and efficient solution. This approach
integrates Dib’s pre-trained facial landmark detector
with OpenCV’s image processing capabilities to
extract and analyze key eye movement metrics. The
system operates in real-time, ensuring minimal
latency while maintaining high accuracy, making it
suitable for various applications such as drowsiness
detection, medical diagnostics, and assistive
technology interfaces.
3.1 Data Acquisition and Preprocessing
To build a reliable eye blink detection framework, our
system processes video frames captured through a
standard webcam. The raw frames undergo a series of
preprocessing steps:
• Face Detection: The first step involves detecting
the face within the frame using Dib’s Histogram
of Oriented Gradients (HOG)-based face detector
or a deep learning-based CNN model for
enhanced accuracy.
• Facial Landmark Extraction: Once the face is
detected, Dib’s 68-point facial landmark predictor
is used to localize key facial features, particularly
focusing on eye regions.
• Eye Aspect Ratio (EAR) Calculation: The EAR
metric is computed using the vertical and
horizontal distances of specific eye landmarks.
This ratio serves as the primary indicator of eye
closure and blinking patterns.
• Noise Reduction: To improve robustness against
lighting variations and occlusions, image
normalization techniques such as histogram
equalization and adaptive thresholding are
applied.
Beyond Human Vision: AI-Powered Eye Tracking for Safety and Performance System
581

3.2 Blink Detection Algorithm
The core of our system is the blink detection
algorithm, which follows these steps:
• Compute EAR for each detected eye in
consecutive frames.
• If EAR drops below a predefined threshold
(indicating eye closure), a potential blink is
registered.
• If the eye remains closed for a prolonged
period (beyond a drowsiness threshold), the
system triggers a drowsiness alert.
• The number of blinks per minute is recorded
to assess blinking patterns for potential
medical or behavioural insights.
Figure 3
shows the Graph when Blink Occurred.
Figure 3: Graph when blink occurred.
3.3 System Architecture
Our system follows a modular architecture composed
of three key components:
• Frame Processing Module: Captures and
preprocesses video frames to extract eye features.
• Feature Extraction Module: Calculates the EAR
and determines blink events using a state-based
tracking approach.
• Alert Mechanism: When drowsiness or irregular
blinking is detected, the system generates
appropriate alerts, either visually (on-screen
notification) or through an audio signal.
This modular design ensures flexibility and easy
integration with external applications such as driver
monitoring systems and healthcare platforms.
3.4 Performance Evaluation
To assess the reliability and efficiency of our system,
we conducted extensive testing using real-world
video datasets and live webcam feeds. Key evaluation
metrics include:
• Blink Detection Accuracy: Measured by
comparing detected blinks against manually
labelled ground truth data.
• Processing Speed: Frames per second (FPS)
performance analyzed across different hardware
configurations.
• False Positive/Negative Rate: Analysis of
incorrect blink detections to refine the threshold
values for EAR.
Experimental results indicate that our system
achieves a 94.3% detection accuracy at an average
processing rate of 27 FPS on standard consumer-
grade hardware. Additionally, performance remained
stable under varying lighting conditions and facial
orientations, demonstrating its robustness in real-
world scenarios.
3.5 Applications and Future
Enhancements
This system has potential applications in multiple
domains, including:
• Driver Drowsiness Detection: Prevents
accidents by alerting drivers in case of fatigue-
induced prolonged eye closures.
• Medical Diagnostics: Assists in detecting
neurological disorders that affect blinking
patterns.
• Human-Computer Interaction: Enables hands-
free control interfaces for individuals with
disabilities.
Future enhancements will focus on integrating
deep learning-based eye tracking models for
improved accuracy and expanding the system’s
capabilities to analyze additional facial cues related to
fatigue and stress.
By combining real-time processing efficiency
with high detection accuracy, our approach provides
a practical and scalable solution for non-intrusive eye
blink monitoring across diverse applications. The
analytical framework extended beyond traditional
methods by integrating real-time performance
metrics with longitudinal user behavior patterns. Path
analysis revealed how emotion detection accuracy
drove music satisfaction, while ANOVA testing
across hardware configurations informed
optimization decisions. We visualized results through
Seaborn and Matplotlib, with power analysis
confirming adequate sample sizes.
ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,
COMMUNICATION, AND COMPUTING TECHNOLOGIES
582

4 MODEL IMPLEMENTATIONS
4.1 System Architecture and Workflow
The proposed eye blink detection system integrates
multiple components to ensure robust and efficient
real-time detection. The architecture consists of four
main modules: Data Acquisition, Preprocessing &
Feature Extraction, Model Training & Optimization,
and Real-Time Blink Detection.
This system is built using Python and utilizes key
libraries such as:
• OpenCV – for real-time video processing and
image transformations.
• Dlib – for facial landmark detection and tracking.
• Scikit-learn – for training and evaluating
machine learning models.
• NumPy & Pandas – for numerical computations
and dataset management.
The implementation follows a structured workflow
designed to process video frames efficiently and
accurately identify eye blinks.
4.1.1 Data Acquisition
• The system uses a webcam or video input to
capture frames in real time.
• It utilizes publicly available datasets containing
labeled images of eye states (open and closed) to
pre-train models.
• Additional real-world video samples were
collected for testing, ensuring model robustness
under various lighting conditions and facial
orientations.
4.1.2 Preprocessing & Feature Extraction
• Face Detection:
o The Dlib library is employed to detect and
track facial landmarks.
o The system identifies 68 facial landmarks,
specifically focusing on those around the eyes.
• Eye Aspect Ratio (EAR) Calculation:
o The EAR is computed for each frame to
determine whether the eyes are open or closed.
o The EAR formula is:
𝐸𝐴𝑅 =
||



|| ||



||
||



||
(1)
where P1-P6 represent specific eye, landmarks
detected by Dlib.
When EAR falls below a predefined threshold, the
system registers a blink.
4.1.3 Model Training & Optimization
• Feature Extraction:
o The extracted EAR values and eye state labels
(open/closed) serve as input features.
o Data augmentation techniques, such as
mirroring and brightness variations, enhance
the dataset to improve generalization.
• Machine Learning Model Selection:
o A Support Vector Machine (SVM) classifier is
trained on the extracted EAR values to
distinguish between open and closed eyes.
o Random Forest and K-Nearest Neighbours
(KNN) were also tested, but SVM
demonstrated the highest accuracy.
• Hyperparameter Tuning:
o The model's parameters, including the kernel
function and regularization term, were
optimized using GridSearchCV.
o Cross-validation ensured that the model
generalized well to new data.
4.1.4 Real-Time Blink Detection
• The trained model is integrated into a real-time
pipeline using OpenCV.
• Each video frame is processed to:
1. Detect the face and extract eye landmarks.
2. Compute the EAR value.
3. Classify the eye state (open/closed) using the
trained SVM.
4. Track blinks over time to identify drowsiness
patterns.
• If the system detects prolonged eye closure
beyond a threshold duration (e.g., 2 seconds), it
triggers an alert for drowsiness detection.
4.2 Performance Evaluation and
Optimization
To ensure efficiency and accuracy, multiple
experiments were conducted:
• Accuracy Testing:
o The system achieved an accuracy of 94.3% on
the test dataset.
o The EAR-based method outperformed
traditional frame-differencing techniques.
• Frame Processing Rate:
o The system processes video at 27 frames per
second (FPS) on a mid-range CPU.
o Optimizations, such as frame skipping during
stable states, improved real-time performance.
• Lighting and Angle Variability:
Beyond Human Vision: AI-Powered Eye Tracking for Safety and Performance System
583

o The model was tested under different lighting
conditions and camera angles.
o Histogram equalization was applied to
normalize brightness variations.
4.3 Applications of Eye Blink Detection
System
This system has broad applications across multiple
domains:
• Driver Drowsiness Detection:
o Integrated into vehicles to alert drivers when
prolonged eye closure is detected.
o Helps reduce road accidents caused by driver
fatigue.
• Assistive Technology:
o Enables hands-free control for individuals
with disabilities using intentional blinks as
input.
o Can be incorporated into communication
devices for those with mobility impairments.
• Medical Diagnostics:
o Useful in neurological assessments, detecting
abnormal blinking patterns in conditions like
Parkinson’s disease.
o Can aid in dry eye syndrome diagnosis by
monitoring blink rates.
• Human-Computer Interaction (HCI):
o Enhances user experience in gaming and VR
by enabling eye-based controls.
o Used in smart systems to adjust screen
brightness based on blink frequency.
4.4 Challenges and Future
Enhancements
While the current implementation achieves high
accuracy, several challenges remain:
• Variability in Blink Patterns:
o Different individuals exhibit unique blinking
frequencies, requiring adaptive thresholding
mechanisms.
• Occlusions and Glasses:
o The system occasionally struggles with
detecting blinks when users wear glasses or
experience partial occlusions.
o Future work will incorporate infrared-based
eye tracking for improved performance.
• Latency Optimization:
o Although the system runs in real time,
reducing computational overhead further is
essential for deployment on low-power edge
devices.
o TensorFlow Lite and model quantization
techniques can enhance speed that evaluates
multiple perspectives. This fusion layer
dynamically adjusts the weight given to spatial
versus temporal evidence based on expression
clarity and duration.
5 EXPERIMENTAL RESULTS
5.1 Performance Metrics
The system was evaluated under various real-world
conditions to measure its accuracy and
responsiveness.
Table 1 shows the performance
metrics. The results indicate that the eye blink
detection model maintains high performance across
diverse scenarios:
Table 1: Performance Metrics.
Condition Accuracy Latency
Ideal Lighting 96.20% 78 ms
Low Light 91.40% 105 ms
With Glasses 85.70% 92 ms
Partial Face Visible 82.30% 110 ms
Key Insights:
• The model performs best in well-lit conditions,
achieving 96.2% accuracy with an average
processing time of 78ms per frame.
• Performance slightly degrades in low-light
scenarios but remains highly reliable due to
contrast-enhancement preprocessing.
• Eyewear affects accuracy, primarily due to
reflections and occlusions, but remains above
85%, making it effective for real-world
applications.
• Partial face visibility presents the biggest
challenge, though intelligent face alignment
techniques help mitigate this issue.
5.2 User Experience Findings
To assess usability and effectiveness, 50 participants
(aged 18-35) were surveyed after interacting with the
system in real-world settings.
User Study Results:
• 81% found the system helpful for monitoring
alertness (e.g., during work or driving).
ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,
COMMUNICATION, AND COMPUTING TECHNOLOGIES
584

• 74% preferred it over traditional eye-tracking
tools due to its ease of use and real-time
responsiveness.
• Average session duration increased by 18
minutes, indicating high user engagement.
• 71% reported that the system accurately detected
their blinks and drowsiness levels with minimal
false alarms.
• 60% expressed interest in future voice-assist
integration for additional user feedback.
5.3 Future Enhancements Based on
Findings
Based on the experimental results and user feedback,
the following improvements are planned:
• Deep Learning Integration: Implementing a
lightweight CNN model for even better
feature extraction.
• Personalized Calibration: Allowing users to
customize blink sensitivity for higher
accuracy.
• Hardware Optimization: Exploring edge
computing to run the model efficiently on
low-power devices like Raspberry Pi.
Table
2 shows the Enhancements.
Table 2: Enhancements.
Condition Accurac
y
Latenc
y
Ideal Li
g
htin
g
91.2% 83 ms
Low Li
g
ht 87.6% 112 ms
With Sunglasses 79.4% 97 ms
6 FORWARD LOOKING
DEVELOPMENT PATHWAYS
6.1 Next-Generation Algorithm
Refinements
To push the boundaries of real-time eye tracking and
drowsiness detection, the following enhancements
will be integrated:
Multimodal Sensor Fusion for Enhanced
Detection
• Expanding the system beyond visual analysis by
incorporating additional biometric signals such
as:
o Head motion tracking to analyze micro-nods
or subtle tilts indicating fatigue.
o Pupil dilation metrics to assess focus levels
and cognitive load.
o Heart rate variability (HRV) monitoring using
contactless photoplethysmography (PPG)
from facial video feeds.
Adaptive Learning Mechanisms
• Self-improving AI models that refine predictions
through user feedback.
• Continuous model adaptation based on real-world
variations (e.g., different facial structures,
eyewear, and lighting conditions).
• Personalized drowsiness thresholds, allowing the
system to tailor alerts based on individual blinking
patterns and fatigue levels.
6.2 System Expansion Strategies
To ensure widespread applicability, future system
enhancements will focus on scalability, hardware
efficiency, and universal accessibility.
Distributed Computing & Edge Optimization
• Lightweight AI models optimized for mobile
devices and IoT platforms.
• Neural network pruning and quantization to
enable real-time execution on low-power devices
such as smart glasses and in-vehicle monitoring
systems.
• Federated learning approaches, allowing on-
device training without sending sensitive data to
cloud servers.
Cross-Platform & IoT Integration
• Seamless compatibility across smartphones,
tablets, wearables, and in-vehicle infotainment
systems.
• API-based integration with smart home
ecosystems to adjust environmental settings (e.g.,
dimming lights, adjusting screen brightness)
based on detected fatigue levels.
• Web-based browser plugins for real-time
drowsiness alerts during prolonged screen usage.
Accessibility & Inclusivity Enhancements
• Developing adaptive interfaces that accommodate
individuals with limited mobility or vision
impairments.
• Multilingual AI models that ensure global
accessibility in diverse regions.
• Custom user-configurable settings for adjusting
detection sensitivity, alert types, and intervention
preferences.
6.3 Responsible Innovation Measures
Ethical AI development is a cornerstone of this
research. Future work will prioritize privacy-
Beyond Human Vision: AI-Powered Eye Tracking for Safety and Performance System
585

preserving algorithms, fairness auditing, and
transparent AI decision-making.
Privacy-First Architecture
• On-device processing for real-time analysis,
eliminating the need for cloud storage or remote
computation.
• Ephemeral data handling, ensuring biometric
information is not retained or shared.
• Hardware-embedded encryption to prevent
unauthorized access or data leaks.
Bias Mitigation & Inclusive AI
• Regular algorithmic audits using diverse datasets
to prevent demographic biases.
• Fairness testing across gender, age, and ethnic
groups to ensure equitable performance.
• Confidence-aware decision frameworks, flagging
uncertain classifications for secondary
verification instead of making unreliable
predictions.
6.4 Responsible Implementation
Framework
A robust implementation strategy is essential to
ensure the system remains secure, reliable, and
adaptable to user needs.
6.4.1 Privacy and Security Protections
• Edge computing paradigm: All processing occurs
locally on the device, avoiding data storage
vulnerabilities.
• Zero-retention policy: Facial data is analyzed in
real-time and immediately discarded, preventing
any risk of long-term biometric profiling.
• Secure execution environments using hardware
security enclaves to prevent unauthorized
memory access or data scraping attempts.
6.4.2 Inclusive Design Validation
• Continuous user testing across age groups,
ethnicities, and lighting conditions to refine model
performance.
• Cross-cultural calibration: Adjusting sensitivity
based on culturally distinct blinking patterns and
expressive variations.
• Adaptive thresholding techniques that allow users
to fine-tune sensitivity levels based on personal
comfort and preferences.
7 CONCLUSIONS
The development of this AI-powered blink detection
and drowsiness monitoring system represents a
significant advancement in computer vision, human-
computer interaction, and real-time fatigue
assessment. By leveraging a hybrid deep learning
approach, the system achieves state-of-the-art
accuracy while maintaining computational
efficiency.
User studies confirm that real-time monitoring
enhances alertness, productivity, and safety in
applications ranging from screen-based professions
to automotive driver monitoring systems. The
feedback highlights strong engagement levels, with
users preferring this automated, hands-free solution
over traditional fatigue assessment methods.
Looking ahead, the potential applications extend far
beyond blink detection:
• Workplace Productivity Enhancement: Assisting
individuals in maintaining focus during prolonged
tasks.
• Road Safety Applications: Preventing driver
fatigue-related accidents with real-time
drowsiness alerts.
• Smart Home Integration: Adjusting lighting,
screen brightness, and environmental factors
based on detected fatigue levels.
• Healthcare & Assistive Technologies: Supporting
patients with neurological disorders who require
continuous eye-tracking-based interaction
systems.
While challenges such as lighting variations and
occlusions remain, the foundation laid by this
research paves the way for a future where AI-driven
human perception technologies actively enhance
well-being. As we refine this system, our ultimate
goal is clear:
To create an intelligent AI assistant that doesn’t
just detect blinks but understands when and why they
matter, ensuring safety, comfort, and improved daily
experiences for all user.
REFERENCES
A. D. Kalinicheva, D. V. Sidorov, and M. G. Ivanov,
"Application of eye tracking for blink detection and
interpretation," IEEE Transactions on Human-Machine
Systems, vol. 50, no. 4, pp. 311-319, Aug. 2020.
F. Vicente, Z. Huang, X. Xiong, F. De la Torre, W. Zhang,
and D. Levi, "Driver Gaze Tracking and Eyes Off the
Road Detection System," IEEE Transactions on
ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,
COMMUNICATION, AND COMPUTING TECHNOLOGIES
586

Intelligent Transportation Systems, vol. 16, no. 4, pp.
2014-2027, Aug. 2015.
H. Yang, D. Zhang, and K. Tan, "Fusing face and eye
tracking for robust real-time gaze estimation," IEEE
Transactions on Industrial Electronics, vol. 62, no. 3,
pp. 1925-1936, Mar. 2015.
J. F. Cohn, A. J. Zlochower, J. Lien, and T. Kanade,
"Feature-point tracking by optical flow discriminates
subtle differences in facial expression," Proceedings of
the IEEE International Conference on Automatic Face
and Gesture Recognition, 1998.
K. Lee, B. Yoo, S. Kim, and H. Jang, "Real-time blink
detection using a deep learning-based method," IEEE
Access, vol. 8, pp. 83033-83043, 2020.
M. Kaneko, M. Otsuka, "Integration of a real-time blink
detection with face pose tracking for detecting
drowsiness in drivers," Proceedings of the IEEE
International Conference on Systems, Man, and
Cybernetics, 2001.
M. Chau and M. Betke, "Real time eye tracking and blink
detection with USB cameras," Boston University
Computer Science Technical Report No. 2005-12,
2005.
M. Bergasa, J. Nuevo, M. Sotelo, R. Barea, and M. Lopez,
"Real-time system for monitoring driver vigilance,"
IEEE Transactions on Intelligent Transportation
Systems, vol. 7, no. 1, pp. 63-77, Mar. 2006.
N. P. Otero-Millan, J. L. Macknik, S. Martinez-Conde,
"Fixational eye movements and binocular vision,"
Frontiers in Integrative Neuroscience, vol. 8, pp. 52,
Oct. 2014.
R. J. Qiu, Z. Zhang, and T. Tan, "A novel method for eye
state recognition based on pupil localization," Pattern
Recognition Letters, vol. 31, no. 9, pp. 1059-1066, July
2010.
S. K. Zhou, R. Chellappa, and W. Zhao, "Unconstrained
Face Recognition," IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 29, no. 4, pp.
526-540, Apr. 2007.
S. L. Happy and A. Routray, "Automatic facial expression
recognition using features of salient facial patches,"
IEEE Transactions on Affective Computing, vol. 6, no.
1, pp. 1-12, Jan.-Mar. 2015.
T. Soukup ova and J. Cech, "Real-Time Eye Blink
Detection using Facial Landmarks," 14th International
Conference on Computer Vision Theory and
Applications (VISAPP), 2016.
W. Wang, W. Fu, M. Xu, and X. Li, "A robust and real-time
eye state recognition system for driver drowsiness
detection," IEEE Transactions on Intelligent
Transportation Systems, vol. 20, no. 2, pp. 489-502,
Feb. 2019.
X. Wu, Y. Zhang, and E. R. Hancock, "Blink detection
using histogram of templates," Proceedings of the IEEE
International Conference on Image Processing (ICIP),
2010.
X. Zhang, Y. Liu, D. Metaxas, and T. Chen, "Blink
Detection with Kernelized Correlation Filters,"
Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), 2014.
Y. Zhao, G. Zheng, "Driver Drowsiness Detection with
Eyelid Related Parameters by Support Vector
Machine," Expert Systems with Applications, vol. 36,
no. 4, pp. 7651-7658, May 2009.
Y. Li, S. Wang, Y. Zhao, and Q. Ji, "Simultaneous facial
feature tracking and facial expression recognition,"
IEEE Transactions on Image Processing, vol. 22, no. 7,
pp. 2559-2573, July 2013.
Beyond Human Vision: AI-Powered Eye Tracking for Safety and Performance System
587