Dynamic Emerging Pathways in Entrance and Exit Detection:
Integrating Deep Learning and Mathematical Modeling
LOAI ABDALLAH, MUTLAQ HIJAZI, MURAD MUSTAFA BADARNA
Department of Information Systems,
The Max Stern Yezreel Valley Academic College,
D.N. Emek Yezreel, 1930600,
ISRAEL
Abstract: - Entrance and exit event detection in dynamic environments has a lot of real-world applications in
security, crowd management, and retail analytics. Traditional methods used for this problem, namely Line
Partition and Bounding Box Diameter methods often struggle in complex scenarios that contain less predictable
movement patterns of individuals. This paper proposes a model that integrates deep learning-based object
detection and tracking techniques with linear regression to enhance the overall performance of enter and exit
detection in static and dynamic environments. This approach captures the movement patterns using advanced
object detection and tracking algorithms, enabling the extraction of y-coordinate variations from bounding box
centers which are used to calculate the tangent of the linear regression equation and determine if the event is
entrance or exit. Experimentations were conducted on 132 video sequences and show the superiority of our
approach over the traditional methods achieving an overall accuracy of 86.36% and an F1-score of 0.86. These
results demonstrate the high efficiency of this approach to accurately detect entrance and exit events, making it
highly reliable and applicable to this problem. This research contributes to computer vision by integrating
object detection and tracking algorithms with linear regression offering a solution for enhancing entrance and
exit events detection in dynamic environments.
Key-Words: - Deep Learning, Mathematical Modeling, Linear Regression, Object Detection, Behavioral
Analysis, Computer Vision.
Received: April 11, 2024. Revised: October 13, 2024. Accepted: November 15, 2024. Published: December 16, 2024.
1 Introduction
Detection of entrance and exit events can be useful
and even critical in domains such as security, crowd
management, and retail analytics. For example, it
can be very useful to detect the entrance of people to
a certain area for security or safety-related causes.
This type of application is helpful in decision-
making, and resource allocation, and therefore it is
crucial to make accurate detections. This research
proposes an activity detection model that focuses on
detecting the entrance and exit events of individuals
by integrating two main components: deep learning-
based object detection and mathematical modeling
using linear regression. Object Detection is a task
that can be solved using deep learning, YOLO is
such a framework that proved its efficiency and
accuracy in localizing objects with high accuracy
and in real time, [1]. When it comes to mathematical
modeling, Linear regression has long been
employed to predict patterns and trends within
datasets, [2]. This paper contributes new knowledge
to the field of computer vision by integrating deep
learning with mathematical modeling to tackle the
difficulties in detecting entrance and exit events.
Through this combination, we seek to achieve two
main objectives:
1. Deep learning-based object detection
techniques perform more precise localization and
tracking of individuals within complex scenes, using
a reliable algorithm to track individuals leads to
better detection of entrance and exit events.
2. Traditional methods often struggle in dynamic
real-world environments with occlusions, varying
lighting conditions, and intricate movement
patterns, [3]. We aim to develop a method that can
robustly operate in such environments.
To the best of our knowledge, most of the
research that was conducted on dynamic
environments did not consider the activity detection
of individuals' entry and exit. Instead, most of the
research focused on other real-world problems such
as traffic direction detection.
While there have been efforts such as [4], [5] that
focus on automating entrance and exit surveillance
in areas where cameras are restricted for privacy
reasons, like washrooms and changing rooms, their
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
331
Volume 23, 2024
approach uses video frames from entrance areas,
analyzing RGB color histogram variations to detect
entrance and exit events. This method encounters
difficulties in environments with significant lighting
changes or where individuals wear similarly colored
clothing, which can result in misclassifications or
false detections. By selecting specific grids in the
camera view and employing temporal analysis, the
research aims to confirm and classify events such as
Entry, Exit, or Miscellaneous. The method relies on
continuous learning for grid selection in dynamic
environments limits its ability to adapt to changing
scenes or crowded spaces.
Studies akin to our work can be observed within
the employment of deep learning for detecting
moving violations in vehicular scenarios, [6], [7].
Intelligent Transportation System (ITS) proposed in
[8] detects vehicles that travel in the wrong-way
direction and notifies the monitoring center.
Although this system achieves high accuracy, it
might struggle in dynamic traffic situations and in
environments with complex road layouts affecting
its real-time performance. Another study proposed
in [9] combines YOLO for object detection and
Gradient Boosting Machine to prioritize emergency
vehicles in a faster and more accurate manner by
addressing issues related to shadow problems, and
real-time execution of existing models. Although it
achieves high accuracy, it might struggle to
accurately identify vehicles, especially emergency
vehicles in complex environments with more
difficult conditions like challenging weather
patterns with limited visibility. The system proposed
in [10] detects vehicles driving against the traffic
flow by using the YOLO framework for vehicle
detection, and centroid tracking within a defined
region of interest. This system’s drawbacks might
be in highly congested traffic scenarios or where
vehicles frequently change lanes which pose
challenges for the system and potentially leads to
false positives or missed detections of wrong-way
vehicles. Our research tackles the problem of
detecting the entrance and exit events of individual
which differs significantly from detecting vehicles
that are traveling in wrong directions, and that’s due
to the different environments of each task where the
movement patterns and its predictability are
different. Individual entrance and exit events
detection focuses on identifying and monitoring
movements within specific zones, such as entryways
or restricted areas, and analyzing the flow of
individuals entering or exiting these spaces. This
task requires tracking and differentiating between
various directional movements within a controlled
environment. On the other hand, detection of
wrong-way travel typically involves identifying
vehicles or pedestrians moving against the
designated traffic flow on roads or pathways.
When comparing our method to others, a high
accuracy of 91.98% in wrong-way drivers’ detection
was achieved in [8], which works great for
structured environments like roads where
the movement of the tracked objects is predictable.
However, our method is built for the movement of
people, which are often less predictable
environments. While the approach in [11] also did
well with vehicles using PTZ cameras, these
approaches don’t quite handle the complexity of
human movement like ours does. The method in [4]
relied on color variations, but that falls short when
lighting changes or people wear similar colors. On
the other hand, the approaches in [6] and [12] are
strong in vehicle tracking, but they don’t adapt as
well to the intricate, varied human activity that our
method excels in.
The key difference lies in the context and nature
of the movement being observed. While both
involve tracking movement patterns, dynamic
entrance and exit detection focuses on environments
where movement patterns of tracked objects are less
predictable with typically involve lower risk if an
event is misclassified, whereas wrong-way detection
deals with more predictable movement patterns, like
those on roads and pathways, where
misclassification can pose significant risks.
Finally, our study distinguishes itself from
previous research by introducing its innovative
approach that focuses on activity detection by
combining object detection and linear regression
techniques to tackle the challenges of dynamic
entrance and exit detection challenges. In contrast to
traditional methods that often struggle in dynamic
environments by applying linear regression to the
detected individual’s coordinates. Our method
achieves higher accuracy in differentiating entrance
and exit events, making it particularly well-suited
for dynamic environments.
In this paper, we first outline the methodology
used in our study, then, we present the results and
analysis of our findings. Subsequently, we discuss
the implications of our results and their significance
in the broader context. Finally, we conclude with
suggestions for future research directions.
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
332
Volume 23, 2024
2 Methodology Enhanced Entrance-
Exit Detection using Object
Detection and Linear Regression
The methodology of our approach integrates deep
learning-based object detection and tracking and
linear regression techniques to enhance entrance and
exit detection accuracy within dynamic
environments. Our approach includes data
collection, cleanup, analysis, and decision-making,
with each step helping to enhance our entrance and
exit detection.
2.1 Data Acquisition and Preprocessing
The initial step of our approach involves acquiring
spatial information from video frames using
the YOLOv5 object detection model, individuals are
localized within the scene at each frame producing a
series of bounding boxes of individual locations.
In the next step, the StrongSORT tracking
algorithm [6] is applied to keep consistent object
tracking across frames, resulting in a set of
trajectories, each characterized by a sequence of
bounding box coordinates over time which is fed to
a windowing approach to extract movement patterns
effectively. In the windowing approach, sequences
of n-consecutive frames are selected, and for each
frame, the corresponding y-coordinates of the
tracked object centers are recorded. This windowed
dataset captures localized movement tendencies,
enhancing the granularity of our subsequent
analysis.
2.2 Feature Extraction and Linear
Regression
From the acquired y-coordinate data, pairwise
distances are computed between all points within
each window. The selection of points with minimum
pairwise distances aims to identify coherent
movement trajectories while filtering out potential
noise and outliers. With the selected y-coordinate
points, a linear regression model is constructed. This
model captures the underlying linear relationship
between time and vertical displacement.
Specifically, we express the direction of movements
as a function of time (), coordinates
() using the equation:
󰇛󰇜
󰇛󰇜󰇛󰇜󰇟 
󰇛󰇜󰇠 (1)
describes the tangent of the linear regression line
which is the rate of vertical movement. We compute
the distance between the coordinates and eliminate
closes values to make the approach more stable and
to identify coherent trajectories. We achieved that
using a function 󰇛󰇜 that defined the threshold:
󰇛󰇜
 (2)
󰇛󰇜
󰇛󰇜󰇛󰇜󰇟󰇛󰇜
󰇛󰇜
 󰇠
󰇛󰇜󰇛󰇜󰇩󰇛
󰇜󰇧󰇛󰇛󰇜󰇜

󰇼
󰇼󰇨󰇪 (3)
where denotes the threshold.
Our algorithm steps are described as follows:
1. YOLO5 is used for object detection and
localization.
2. A StrongSORT tracker used to track objects
across video stream frames.
3. Sequential frames are grouped in windows
of frames each for localized data
acquisition.
4. Calculate pairwise distances between y-
coordinates to identify coherent trajectories.
5. Build linear regression models for selected
y-coordinate data in order to capture linear
relationships.
6. Extract the tangent '' from the linear
regression equation.
7. Apply threshold to '' for precise
entrance/exit identification.
3 Experiments and Results
To rigorously evaluate the effectiveness of our
proposed approach for entrance and exit detection,
we conducted a series of comprehensive
experiments encompassing both quantitative
assessments and qualitative analysis. These
experiments aimed to ascertain the accuracy,
robustness, and real-world applicability of our
methodology, showcasing its potential to address
challenges such as occlusions, camera locations,
filming angles, tracked objects' distance from
the camera, and complex movement patterns.
3.1 Experiments Setup
We employed a diverse dataset comprising video
sequences captured in dynamic environments (i.e.,
various camera locations, altitude, and the angle at
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
333
Volume 23, 2024
which the video sequences were captured). In
addition, the video sequences include corridors,
entrances, and exits. The dataset encompassed
scenarios characterized by camera view restrictions
and multi-directional movement patterns, thereby
simulating real-world complexities.
For the experiment purposes, we captured video
sequences in an unoccupied space that we
segmented into three rows: front, center, and back in
relation to the entrance/exit location (Figure 1). We
further subdivided each row into 3 locations: left,
center, and right in relation to the camera’s location.
In addition, the video sequences captured in each
location were characterized by three altitudes: high,
middle, and low in relation to the ground. We also
captured video sequences in which the camera’s
capturing angle was subtly adjusted at each location
at the altitude which was categorized as high. For
simplicity, we divided the various camera locations
into a 3x3 grid, where each cell has three video
sequences, and the camera is either located at a
high, middle, or low altitude, in addition to
the video sequence that the camera’s capturing
angle was subtly adjusted, the angle adjustments
occur only on the right and left columns of the grid.
We have 27 locations, 3 videos were captured at
each location, in addition to 6 videos with capturing
angle adjustment. To simulate real-world scenarios,
we applied two entrance, and two exit scenarios that
simulate real-world scenarios: straight movement of
the tracked individual towards or against the
location of the camera which makes in a total of 66
videos of entrances, and 66 videos of exits, 132 in
total that was tested.
Fig. 1: Camera Location Grid: we captured at every
camera location the entrance and exit video at 3
different altitude levels: high, middle, and low, in
addition to angle-adjusted videos at left and right
locations at level high
3.1.1 Video Sequence Characteristics
Our research included 133 videos where each video
is an entrance scenario or an exit scenario, our
dataset includes the following statistics:
1. The videos were captured with a constant frame
rate of 20 frames per second (fps).
2. The videos range from 40-frame videos to longer
ones consisting of 180 frames, which makes it
range from 2 seconds videos up to 9 seconds
videos, while the mean duration is approximately
4.84 seconds.
3. The videos were captured with a frame width of
704 pixels and a height of 576 pixels.
3.1.2 Compared Methods
Line Partition Method: This approach is based on a
scene division strategy using a designated line, the
line was horizontally at 60% of the width. If tracked
individuals are found above the line at the end of a
video sequence, then the scenario is interpreted as
an entrance, while their presence below the line
signifies an exit. While this method is simple and
easy to implement, it heavily depends on the exact
placement of the partition line and fails to consider
the specific movement patterns on either side of the
line. In contrast, our method leverages tangent
values for nuanced direction detection, allowing for
greater accuracy in complex scenarios.
Bounding Box Diameter Method: In this
method, we captured the dimensions of the
bounding box obtained from YOLOv5 object
detection over each frame in the video sequence.
Next, a window approach is applied to calculate the
mean of these bounding box diameter values over
20 consecutive frames. At each frame, the mean
value of the bounding box diameters in the frame
window is updated. If the new mean is smaller than
the previous one, the method suggests that the event
is an exit, whereas if it is larger, it indicates an
entrance. The scenario is classified based on the last
two means, if the last one is larger than its
predecessor then the scenario is classified as exit,
otherwise, it is classified as entrance.
3.2 Experiments
We have evaluated the performance of our proposed
approach by comparing it with two alternative
methods mentioned in the previous section: the Line
Partition, and the Bounding Box Diameter methods.
To evaluate our method, we formulated
performance metrics to measure the accuracy of our
entrance and exit detection method. We use metrics
such as F1-score and accuracy to gauge the method's
ability to accurately detect entrances and exits in
comparison to ground truth annotations. We
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
334
Volume 23, 2024
calculate these metrics across a range of dynamic
environment scenarios to assess the method's
robustness and consistency. Each method’s ability
was evaluated to detect entrances against its ability
to detect exits, in addition to the performance of
each method at each camera location.
Fig. 2: Linear Regression-based method’s accuracy
in detecting entrances vs. exits showing accuracy
comparison in a straight motion
Fig. 3: Linear Regression-based method’s accuracy
in detecting entrances vs. exits showing accuracy
inside motion
Fig. 4: Line Partition method’s accuracy in detecting
entrances vs. exits showing overall accuracy scores
for our method
As part of our experiments, the performance of
each method was compared to detect entrance and
exit while the tracked individual’s movement is
vertical, meaning it is moving towards the location
of the camera in the entrance or against the location
of the camera in the exit. As can be seen in Figure 2,
Figure 3 and Figure 4. The accuracy score achieved
by our method is 90.9%, and the f1-score is 0.9 in
a straight motion, while in side-motion it achieved
81.8% accuracy and f1-score of 0.81. The overall
accuracy score of our method has achieved 86.36%,
and an f1-score of 0.86.
Figure 5, Figure.6 and Figure 7 show the results
of the Bounding Box Diameter method. The
accuracy score achieved is 71.21%, and the f1-score
is 0.68 in a straight motion, while in side-motion it
achieved 56.72% accuracy and an f1-score of 0.5.
The overall accuracy score that this method has
achieved is 63.91%, and the f1-score of 0.59.
Fig. 5: Bounding Box Diameter method’s accuracy
in detecting entrances vs. exits showing accuracy
comparison in straight motion
Fig. 6: Bounding Box Diameter method’s accuracy
in detecting entrances vs. exits showing accuracy on
side motion
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
335
Volume 23, 2024
Fig. 7: Bounding Box Diameter method’s accuracy
in detecting entrances vs. exits showing overall
accuracy scores for our method
Next, a comparison between all the methods
was conducted to evaluate each method’s exit
detection, entrance detection, and overall
performance against each other. As can be seen in
Figure 8 and Figure 9, the linear regression-based
algorithm outperforms the other two methods in
both metrics F1-score and Accuracy. This is to be
expected because the linear regression-based
algorithm considers better the dynamic nature of
scenarios captured by the tested videos.
Fig. 8: F1-Score comparison between methods
Fig. 9: Accuracy Comparison for different methods
3.3 Qualitative Analysis
In addition to quantitative measures, we conducted a
qualitative analysis by visualizing the trajectory
predictions generated by our linear regression
algorithm. Through these visualizations, we
discerned that our approach successfully captured
intricate movement patterns, effectively
distinguishing between entrance and exit directions.
Figure 10, Figure 11 and Figure 12 visually depict
how each of the discussed methods has been applied
to scenarios involving entrances and exits.
This observation was particularly evident in
scenarios featuring multiple individuals and
occlusions, where traditional methods often faltered.
We observe that each method is better at one of the
scenarios examined. As expected, our method has
achieved higher accuracy scores in detecting exit
events than in entrance events, while the other two
methods had higher accuracy scores in detecting
entrance events than exit. This is because the
duration time of the exit activity is shorter than the
entering activity in the tested videos, making the
detection difficult for the other methods to detect the
individual in all required video frames. As a result,
other methods are not able to capture the exit
activity. Our method tracks the individual in the
video frames and based on his movement direction
it detects the correct activity.
In all metrics, and across all scenarios, our
method performs best, followed by the Bounding
Box Diameter method, while the Line Partition
method falls behind. The goal of the experiments
that we conducted was to examine the ability of
different methods to detect entrances and exits with
minimum human intervention by applying the same
configuration of each method on different camera
locations, altitudes, and angles. This constraint is a
main cause that makes Line Partition method
struggle in many different locations, altitudes, and
angles.
Fig. 10: On the left Line Partition method on
straight motion entrance scenario in a location with
low altitude. On the right Line Partition method exit
scenario in a location with high altitude
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
336
Volume 23, 2024
4 Discussion
In terms of the Line-Partition approach, although it
is a widely used method for activity detection, it
exhibits notable limitations. Firstly, it relies on
manual human intervention to define entrance and
exit zones within the monitored area, where if the
tracked individuals are present in an exit zone the
event is considered to be an exit, otherwise it’s
the entrance. This logic may not seamlessly
integrate into dynamic monitoring environments and
might necessitate the intervention of humans at each
change of environment settings. Without this human
intervention, and similar to the findings in [5], the
method struggles to accurately distinguish true
entrances and exits in environments with variability
in movements, potentially leading to false
detections. Secondly, the division of the monitored
area into entrance and exit areas leads to
disregarding important information which might be
crucial for activity detection and that’s because it
does not track the exact movement of the
individuals inside the predefined area. A better
approach to tackle this problem is the Bounding Box
Diameter approach, which is not dependent on
human intervention and achieves better results in
dynamic environments. The main disadvantage of
this approach is that it is not directly motion-based,
instead, it relies on the bounding box dimensions
resulting by object detection which can sometimes
exhibit sudden and unexpected changes that do not
accurately represent the motion of the tracked
objects making it not highly stable and is likely to
face challenges when the movement patterns of
individuals are less predictable. The results we
conducted in this research demonstrate the
effectiveness and the dynamism of our approach by
achieving extraordinary results in automatic
entrance and exit events detection in dynamic
environments where the movement patterns of
individuals are less predictable and higher
adaptability to dynamic environments than prior
studies such as [8].
5 Conclusions
In this paper, we introduce a novel approach to
entrance and exit detection in dynamic
environments by integrating deep learning-based
object detection with mathematical linear
regression. Commonly used methods that address
this problem, namely the Line-Based method and
the Bounding Box Diameter method, struggle in
dynamic environments where the movement
patterns of the monitored individuals are less
predictable. Our approach demonstrates high
adaptability to various environment settings with no
manual human intervention while achieving
superior results over commonly used methods with
an overall accuracy of 86.36%, and an F1-score of
0.86, making it a suitable option for complex real-
world environments that introduce additional factors
such as occlusions, varying lighting conditions, and
complex movement patterns which challenge both
existing methods and systems proposed by other
studies which are more suitable for structured
environments such as roadways where the
movement patterns of the tracked objects are more
predictable than the movement patterns of human
individuals. In addition, the adaptability of our
approach while maintaining good performance
makes our approach well-suited for applications in
domains such as security, crowd management, and
retail analytics, where precise monitoring of human
movement is critical. Future research could explore
the adaptation of this method to other areas,
including behavior analysis, vehicular movement
analysis, or real-time monitoring in public spaces, to
further validate its effectiveness and versatility. We
also show that the integration of deep learning and
mathematical modeling holds significant potential
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
337
Volume 23, 2024
for advancing the field of computer vision and
developing more robust and accurate detection
systems. This work lays the foundation for further
innovations in intelligent monitoring technologies,
paving the way for enhanced applications in
increasingly complex environments.
Declaration of Generative AI and AI-assisted
Technologies in the Writing Process
During the preparation of this work the authors used
ChatGPT developed by OpenAI in order to perform
minor revisions to enhance readability if certain
phrases. After using this tool/service, the authors
reviewed and edited the content as needed and take
full responsibility for the content of the publication.
References:
[1] Ultralytics, YOLOv5: "A state-of-the-art real-
time object detection system", 2021.
[2] Hocking, R. R. "Developments in Linear
Regression Methodology: 1959-1982."
Technometrics, vol. 25, no. 3, 1983, pp. 219–
230. DOI: https://doi.org/10.2307/1268603.
Accessed 21 Dec. 2023.
[3] Majumder, M., & Wilmot, C. "Automated
Vehicle Counting from Pre-Recorded Video
Using You Only Look Once (YOLO) Object
Detection Model." J. Imaging, vol. 9, no. 7,
2023, p. 131. DOI:
https://doi.org/10.3390/jimaging9070131.
[4] Vinay Kumar V. & P. Nagabhushan.
"Monitoring of people entering and exiting
private areas using Computer Vision." arXiv,
2019. DOI:
https://doi.org/10.48550/arXiv.1908.00716.
[5] Vinay Kumar V., & P. Nagabhushan. "Entry-
Exit event detection from video frames."
International Journal of Computer Sciences
and Engineering, vol. 6, no. 2, 2018, pp. 112-
118.
http://dx.doi.org/10.26438/ijcse/v6i2.112118.
[6] Yunhao Du, Zhicheng Zhao, Yang Song,
Yanyun Zhao, Fei Su, Tao Gong, & Hongying
Meng. "StrongSORT: Make DeepSORT Great
Again." arXiv, 2023. DOI:
https://doi.org/10.48550/arXiv.2202.13514.
[7] Manasa, A., & Renuka Devi, S. M. "An
Enhanced Real-Time System for Wrong-Way
and Over Speed Violation Detection Using
Deep Learning." Fourth International
Conference on Image Processing and Capsule
Networks, 2023, pp. 309-322. DOI:
https://doi.org/10.1007/978-981-99-7093-
3_21.
[8] Usmankhujaev, Saidasul, Shokhrukh
Baydadaev, & Kwon Jang Woo. "Real-time,
deep learning based wrong direction
detection." Applied Sciences, 2020. DOI:
https://doi.org/10.3390/app10072453.
[9] Vikruthi, S., Archana, M., & Tanguturi, R. C.
"A Novel Framework for Vehicle Detection
and Classification Using Enhanced YOLO-v7
and GBM to Prioritize Emergency Vehicle."
International Journal of Intelligent Systems
and Applications in Engineering, vol. 12, no.
1S, 2023, pp. 302-312.
[10] Bharadwaj, R., Billade, A., Chenna, S.,
Chandrapatle, A., & Chinchalpalle, G.
"Wrong Way Vehicle Detection in Single and
Double Lane." IJRITCC, vol. 11, no. 6S,
2023, pp. 457-462. DOI:
https://doi.org/10.17762/ijritcc.v11i6s.6953.
[11] Haghighat, A., & Sharma, A. "A computer
vision-based deep learning model to detect
wrong-way driving using pan–tilt–zoom
traffic cameras." Computer-Aided Civil and
Infrastructure Engineering, vol. 38, no. 1,
2023, pp. 119–132. DOI:
https://doi.org/10.1111/mice.12819.
[12] Suttiponpisarn, P., Charnsripinyo, C.,
Usanavasin, S., & Nakahara, H. "An
Autonomous Framework for Real-Time
Wrong-Way Driving Vehicle Detection from
Closed-Circuit Televisions." Sustainability,
vol. 14, no. 16, 2022. DOI:
https://doi.org/10.3390/su141610232.
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed in the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts of interest to declare.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.37
Loai Abdallah, Mutlaq Hijazi,
Murad Mustafa Badarna
E-ISSN: 2224-2678
338
Volume 23, 2024