Volume 04 Issue 04-2024
74
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
04
Pages:
74-78
SJIF
I
MPACT
FACTOR
(2022:
5.636
)
(2023:
6.741
)
(2024:
7.874
)
OCLC
–
1368736135
A
BSTRACT
The exponential growth of urbanization has led to increasing problems in traffic management, which
require innovative solutions for efficient estimation of traffic flow. Deep learning methods are emerging as
a powerful tool for processing and analyzing satellite images, and are being used for traffic flow estimation.
This paper describes deep learning-based methods for traffic flow estimation using satellite imagery.
K
EYWORDS
Transport Flow, traffic flow, satellite images, urban traffic.
I
NTRODUCTION
Today, urban traffic is a common problem
affecting the quality of life in modern cities.
Accurate estimation of traffic flow is essential for
effective traffic management, route planning and
infrastructure development. Traditional methods
of traffic flow estimation, such as detectors and
surveillance cameras, suffer from limitations such
as high cost, limited coverage, and maintenance
requirements. In contrast, satellite imagery offers
a comprehensive and cost-effective alternative to
monitoring traffic flow over large spatial areas.
With recent advances in deep learning, the ability
to automatically learn relevant features and
patterns from raw data, analyzing satellite
Journal
Website:
http://sciencebring.co
m/index.php/ijasr
Copyright:
Original
content from this work
may be used under the
terms of the creative
commons
attributes
4.0 licence.
Research Article
EVALUATION OF TRANSPORT FLOW USING DEEP LEARNING
METHODS OF SATELLITE IMAGES
Submission Date:
April 13,
2024,
Accepted Date:
April 18, 2024,
Published Date:
April 23, 2024
Crossref doi:
https://doi.org/10.37547/ijasr-04-04-13
Hamrakulov Botirbek
TUIT named after Muhammad al-Khwarizmi, Uzbekistan
Volume 04 Issue 04-2024
75
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
04
Pages:
74-78
SJIF
I
MPACT
FACTOR
(2022:
5.636
)
(2023:
6.741
)
(2024:
7.874
)
OCLC
–
1368736135
imagery for traffic flow estimation, is currently
being extensively researched.
Deep learning architectures for traffic flow
estimation using satellite imagery.
Several deep learning architectures, including
convolutional neural networks (CNNs), recurrent
neural networks (RNNs), and their variants, have
been used to estimate traffic flow using satellite
imagery. CNNs are widely used for feature
extraction from satellite imagery due to their
ability to efficiently capture spatial hierarchies.
RNNs, specifically Long Short-Term Memory
(LSTM) networks, have been used to model
temporal dependencies in traffic flow data.
Hybrid architectures combining CNNs and RNNs
also provide superior performance in spatial and
temporal data acquisition for more accurate
traffic flow estimation.
Figure 1. Satellite image for traffic flow estimation.
In the case above, the purple colored images show
the parked cars, the blue ones are moving, and the
yellow ones are moving cars.
A comparative analysis of CNN and RNN
Table 1.
Comparative
characteristics
Convolutional Neural Networks
(CNN)
Recurrent Neural Networks
(RNN)
Release features
CNN excels in extracting spatial
features from satellite images.
RNNs aim to capture temporal
correlation in sequential data.
Volume 04 Issue 04-2024
76
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
04
Pages:
74-78
SJIF
I
MPACT
FACTOR
(2022:
5.636
)
(2023:
6.741
)
(2024:
7.874
)
OCLC
–
1368736135
Architecture
It usually consists of
convolutional layers followed by
pooling layers.
Composed of repeating units such as
LSTM or GRU cells.
Strengths
- Effective in extracting spatial
relationships and visual patterns.
Parallel processing for real-time
applications.
- The ability to model temporal
correlations over time. It is suitable
for serial data with long-term
correlations.
Disadvantages
- Limited in modeling temporal
dynamics and long-term
dependencies. It may require
additional mechanisms for
temporal modeling.
- Complexity increases with longer
sequences. Limited in capturing
spatial features.
Data
requirements
Annotated satellite imagery data
sets are relied upon for training.
Requires serial traffic flow data with
timestamps for training.
Complexity of
training
Usually less complex due to
parallel processing of spatial
features.
More complex due to sequential
processing and acquisition of
temporal dependencies.
Ability to
interpret
Can visualize learned features in
convolutional layers.
Interpretation can be difficult due to
complex temporal relationships.
Architectural
examples
VGG, ResNet, DenseNet
LSTM, GRU
Datasets and preprocessing methods.
Volume 04 Issue 04-2024
77
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
04
Pages:
74-78
SJIF
I
MPACT
FACTOR
(2022:
5.636
)
(2023:
6.741
)
(2024:
7.874
)
OCLC
–
1368736135
The availability of diverse and annotated datasets
is essential for training and evaluating deep
learning models for traffic flow estimation.
Several public datasets, such as the SpaceNet
dataset and the DeepGlobe dataset, provide high-
resolution satellite imagery and ground-based
annotations for traffic flow. Preprocessing
techniques such as image enhancement,
normalization, and data augmentation are
commonly used to improve the quality and
diversity of training data, thereby increasing the
robustness and generalizability of deep learning
models.
Various evaluation metrics are used to evaluate
the performance of deep learning models for
traffic flow estimation, including mean absolute
error (MAE), root mean square error (RMSE), and
intersection by union (IoU). These metrics
determine the accuracy, precision, and
consistency of predicted traffic flow compared to
ground truth data, allowing researchers to
effectively evaluate and compare different
models.
Despite the promising results achieved by
methods based on deep learning, there are a
number of problems in the field of traffic flow
estimation using satellite images. Challenges
include the limited amount of labeled data, the
complexity of urban environments, and the
interpretability of deep learning models. Future
research directions should address these
challenges by developing new algorithms,
integrating multimodal data sources, and
exploring advanced deep learning techniques
such as generative adversarial networks (GANs)
and reinforcement learning. possible
C
ONCLUSION
Deep learning-based methods are emerging as
effective tools for traffic flow estimation using
satellite imagery, offering advantages such as
scalability, automation, and cost-effectiveness.
Using the power of deep learning architectures,
datasets, processing techniques, and estimation
metrics, researchers use the latest advances in
traffic flow estimation to contribute to smarter
and more manageable transportation systems.
R
EFERENCES
1.
Antoniou, C.; Barcelo, J.; Brackstone, M.;
Celikoglu, H.B.; Ciuffo, B.; Punzo, V.; Sykes, P.;
Toledo, T.; Vortisch, P.; Wagner, P. Traffic
Simulation: Case for Guidelines. 2014.
Available
online:
https://data.europa.eu/doi/10.2788/11382
(accessed on 8 April 2019).
2.
Salehi, B.; Zhang, Y.; Zhong, M. Automatic
moving vehicles information extraction from
single-pass worldView-2 imagery. IEEE J. Sel.
Top. Appl. Earth Obs. Remote Sens. 2012, 5,
135
–
145. [Google Scholar] [CrossRef]
3.
Satellite Imaging Corporation. IKONOS
Satellite Sensor. 2017a. Available online:
https://www.satimagingcorp.com/satellite-
sensors/ikonos/ (accessed on 18 March
2020).
4.
Satellite Imaging Corporation. QuickBird
Satellite Sensor. 2017b. Available online:
Volume 04 Issue 04-2024
78
International Journal of Advance Scientific Research
(ISSN
–
2750-1396)
VOLUME
04
ISSUE
04
Pages:
74-78
SJIF
I
MPACT
FACTOR
(2022:
5.636
)
(2023:
6.741
)
(2024:
7.874
)
OCLC
–
1368736135
https://www.satimagingcorp.com/satellite-
sensors/quickbird/ (accessed on 18 March
2020).
5.
Mace, E.; Manville, K.; Barbu-McInnis, M.;
Laielli, M.; Klaric, M.; Dooley, S. Overhead
Detection: Beyond 8-bits and RGB. arXiv 2018,
arXiv:1808.02443. [Google Scholar]
6.
Zhihuan, W.; Xiangning, C.; Yongming, G.;
Yuntao, L. Rapid target detection in high
resolution remote sensing images using YOLO
Model. Int. Arch. Photogramm. Remote Sens.
Spat. Inf. Sci.-ISPRS Arch. 2018, 42, 1915
–
1920. [Google Scholar]
