International Journal of Scientific Research in Science and Technology

Available online at : www.ijsrst.com

Print ISSN: 2395-6011 | Online ISSN: 2395-602X doi : https://doi.org/10.32628/IJSRST2411476

Automatic Number Plate Recognition

Swanand Joshi, Pramod Jejure, Chatrasal Jadhav, Vishal Jankar
*Department of Computer Engineering, ZCOER, Pune, Maharashtra, India

ARTICLEINFO ABSTRACT
Automatic license plate recognition (ANPR) systems have become suitable
Article History:
for various applications, including traffic monitoring, law enforcement,
Accepted : 12 Oct 2024 and toll collection. This paper completes the study on automatic license
Published : 28 Oct 2024 plate recognition (ANPR) systems that use advanced imaging technology
and machine learning algorithms to achieve accuracy in license plate
verification and validation. The preparation process is adopted in various
Publication Issue : ways: image acquisition, preprocessing, location plate, character
Volume 11, Issue 5 segmentation, and optical character recognition (OCR). The system, which
Sep-Oct-2024 integrates deep learning models for extraction and classification, performs
better in different environments. Experimental results show that the
Page Number : identification accuracy of the proposed ANPR exceeds 95%, demonstrating
439-448 its potential in practical applications. In addition, this paper also discusses
the problems encountered in ANPR implementation, including changes in
plate design, illumination, and shading, and provides suggestions for future
research to improve stability and efficiency. Keywords: automatic plate
recognition, image processing, machine learning, optical behavior
recognition, deep learning, vehicle tracking, driving license inspection.
Keywords: Automatic License Plate Recognition (ANPR), Image
Acquisition, License Plate Localization, Character Segmentation, Optical
Character Recognition (OCR), Identification Accuracy

I. INTRODUCTION identify the license plate. ANPR systems often use

advanced techniques, including deep learning models,
Automatic license plate recognition (ANPR) is an to increase accuracy and robustness, allowing them to
intelligent device that uses optical character perform well in different situations such as different
recognition (OCR) to recognize and clarify driver lighting, angles, and design licenses. ANPR
license plates from images or video services. The most technology plays an important role in modern traffic
common method is to use a camera to capture the management and law enforcement. Its applications
image and then use various types of image processing are unique and useful, from local vehicle maintenance
to enhance and analyze the data to isolate and to assisting traffic laws. ANPR is widely used for

Copyright © 2024 The Author(s): This is an open-access article distributed under the terms of the Creative 439
Commons Attribution 4.0 International License (CC BY-NC 4.0)
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

reimbursement and enables integration without the vertical projection analysis (helps identify differences
need for physical intervention. In the field of security, between characters) and contour detection (shows the
ANPR helps the police track stolen vehicles, manage shape of characters). Advanced technology can also
parking lots, and conduct search operations. In use deep learning models to accurately predict
addition, with the development of smart city plans, boundary markers, especially when markers are close
technology certification has become an integration or not completely obscured. Good segmentation
technology of transportation, improving the ensures that each character is clearly visible, thus
efficiency of urban transportation and public safety. improving the overall presentation.
This research focuses on traveling and improving the
capabilities of ANPR systems by combining image 1.2. MACHINE LEARNING ALGORITHM
processing and machine learning techniques. Machine learning has revolutionized the field of
automatic license recognition (ANPR) by providing
1.1. IMAGE PROCESSING advanced techniques for processing and interpreting
Photographs are the first important step in the ANPR complex data. Unlike traditional methods that rely on
process, where high-quality images of the vehicle are complex rules, machine learning models learn from
captured using a variety of cameras, including fixed, large datasets, allowing them to improve accuracy
mobile and infrared cameras. The quality of the over time. These algorithms adapt to changes in
captured image affects the accuracy of the next permit design, environment, and closure, making the
processing. Use image enhancement techniques such ANPR system robust and efficient. Integrating
as histogram equalization, contrast adjustment and machine learning into the licensing process not only
noise reduction to develop the specific license. This improves search and recognition capabilities, but also
process helps ensure that the license plate markings speeds up execution, which is important for
are independent and easily identifiable, even in applications such as traffic management and law
difficult conditions such as low light or cloudy enforcement.
weather. The next step after capturing the image is to Feature extraction is a key step in the ANPR machine
apply for a local license, which includes defining and learning workflow because it transforms raw data into
distributing the image area covered by the license. meaningful representations that algorithms can
Many techniques have been used for this task, interpret. Techniques include edge detection, angle
including edge detection, morphological operations, detection, and histogram techniques that help identify
and machine learning techniques. While Edge seeks license plate features. In addition, deep learning
insight into the board’s boundaries, the morphological methods, especially convolutional neural networks
function can adjust the search process by eliminating (CNN), extract hierarchical features from images,
noise and improving the board. In addition, allowing models to capture complex patterns without
convolutional neural networks (CNN) have been being affected by the book. Effective screening
recognized for their ability to learn features directly improves the model’s ability to distinguish characters
from images, leading to stronger localization in many and increases general awareness.
areas. After the License is successfully processed, Classification systems play an important role in
character classification is performed to identify each interpreting the extracted features and determining
character separately. This step is dangerous because it the correct sign of the license. Traditional methods
directly affects the accuracy of the optical resolution include support vector machine (SVM), k-nearest
(OCR) process. The methods include horizontal and neighbor (k-NN), and decision trees, which are

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 440
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

excellent for simple data. However, today’s ANPR pattern matching and recognition as a feature, where
systems use deep learning techniques, primarily CNNs, the uniqueness of each character is compared with the
due to their superior performance in image previous model. However, with the advancement of
classification. These algorithms learn from different machine learning, modern OCR technology
recorded data sets, allowing them to classify increasingly relies on neural networks, especially
characters with high accuracy even in difficult neural networks (CNN). These networks can learn
situations. By combining these algorithms, the license complex patterns and changes in character images,
verification system can perform efficient and effective making them more robust to noise, distortion, and
license verification and pave the way for advanced font changes. Additionally, Recurrent Neural
applications in maintenance and security. Networks (RNN) and Long Short-Term Memory
(LSTM) networks are used to improve the recognition
1.3. OPTICAL RECOGNITION of connected data, such as symbols on license plates.
Optical character recognition (OCR) is an important Performance evaluation of OCR methods in ANPR is
part of automatic license plate recognition (ANPR) important to understand their effectiveness and
and acts as a bridge between processed images and reliability. Common metrics used for evaluation
valuable data. OCR converts the information include recognition accuracy, precision, recall, and F1
contained in a driver's license into machine-readable score. These metrics provide insight into how your
text, which is important for applications such as OCR system performs under different conditions,
vehicle identification, license plate registration, and such as different lighting conditions, occlusions, and
law enforcement. The accuracy of OCR directly font styles. Benchmark datasets created specifically
affects the overall performance of the ANPR system, for ANPR evaluation allow researchers to compare
even a small error in character recognition can cause the performance of different OCR tools. Also, real
the vehicle to malfunction. Therefore, good OCR world testing is important to check how the system
technology is essential to ensure the reliability and performs in realworld situations. Continuous
functionality of ANPR applications. Various improvements in OCR techniques and performance
techniques have been developed for character metrics are critical to advancing ANPR technology
recognition in ANPR systems, from traditional to and ensuring it meets today's vehicle and safety
advanced techniques. Early methods generally use requirements.

II. SYSTEM ARCHITECTURE

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 441
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

The presented architecture shows a general

introduction to the process of using license plate 4. Test license:
recognition from images captured by a camera. The • Color-based segmentation: This step uses the
system has important applications in areas such as specific color of the license plate (usually blue or
traffic management, security monitoring and parking yellow) to distinguish it from the vehicle behind.
lot management. • Edge detection and feature removal: Use
techniques such as Canny edge detection and
Architecture Decomposition Haar-like features to define the boundary of the
1. Input (camera): license plate.
• the system starts with an input, which is usually
a camera that captures real-time images of the 5. License Agreement:
vehicle or images from previously recorded • Bounding box: The system determines the exact
images. A location of the license plate in the image, usually
represented by a bounding box.
2. Pre-Treatment: 6. Character Fragmentation:
• Image enhancement: This step includes • Component analysis: This step separates
techniques such as noise reduction, contrast, and individual characters from segmented license
sharpening to improve image quality and license images.
clarity. • Morphological operations: Use techniques such as
• Image normalization: This step normalizes the erosion and dilation to heal broken symbols.
image according to size, orientation, and lighting
for further processing. 7. Personality:
• Optical Character Recognition (OCR): Using
3. Vehicle inspection: OCR techniques such as pattern matching, neural
• Object detection algorithm: The system uses networks, or support vector machine (SVM), the
object detection tools such as Faster R-CNN, system recognizes characters in each segment.
YOLO, or SSD to identify and locate vehicles in • Machine learning models: Pre-learning models or
the image. training models can be used to improve accuracy.

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 1002
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

8. Output:
• The final output is a certified vehicle certificate
that can be displayed on screen, stored in a file,
or used for further processing such as querying
vehicle information.

Key points and potential for improvement

• Image quality robustness: The system must be

able to handle changes in image quality due to
factors such as lighting, camera angle, and
occlusions.
• Real-time processing: The system must process
images in real time or near real time for the data
to be usable.
• Accuracy: Character recognition accuracy is
important, especially for applications that require
high precision.
• Scalability: The system should be scalable to
handle the size of images and videos.
• Integration with other systems: This system can
be integrated with other systems such as traffic
management or security monitoring

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 443
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

FLOWCHART Driving License Front View: The system first takes an

image of the driving license. It is important that the
license plate is clearly visible and in the correct
orientation.
Working phase: This time work on the operation of
the image. It includes: a Convert RGB images to
grayscale images: Convert colour images to grayscale
images for easy later processing.
Image binarization: Grayscale images are converted to
binary images (black and white) to separate characters
from the background.
JPG to Bit Converter: Convert the finished image to a
byte stream for easy storage and transfer
Development phase: Stock images (JPG): This time,
use stock images to demonstrate and test the system.
License Plate: The system shows the license plate area
in the image.
License segmentation: The license area is separated
from the rest of the image.
Character segmentation: Each character on the
driver's license is separated. Behaviour recognition:
Use techniques such as pattern matching or neural
networks to identify each behaviour.
License Certificate: Provide authentication symbols
along with valid license information to verify their
authenticity.
Final output: A system displays the approved license
plate and can be used in many applications such as
traffic monitoring, parking management or police. a
Key point: Real-time runtime image management
process during the development process before storing
images for training and testing. Image pre-processing
such as grayscale conversion and binarization are
important to realize correct behaviour. The system
relies on image embedding, segmentation,
identification and other technologies to capture and
identify vehicle license plates.
Full process: This flowchart shows the steps involved
in a license certificate designed to extract and
recognize license symbols in images. This system
works in two phases: the operational phase and the
development phase. working hours:

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 444
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

III.LITERATUR SURVEY

Name of Paper Author Year Publisher Objective Methodology Limitation

Real Time Car T.Mustafa & 2024 IEEE This system Conducted Many
Model and M.Karabatak Xplore aims to experiments with challenges in
Plate Detection advantage to different ANPR detecting cars
System By vehicle solutions under and plates in
using Deep identification controlled criteria’s different
Learning accuracy and environmental
processing speed conditions, such
in various traffic as poor lighting
conditions. or occlusion
Research on W.Weihong 2021 IEEE To develop a Development of Difficulties in
Licence Plate &T.Jiaoyang Xplore robust license advanced image recognizing
Recognition in plate processing license plates
complex recognition techniques and under extreme
Environment system capable deep learning conditions, such
of functioning models for effective as heavy
effectively in recognition in motion blur or
complex complex severe
environments environment occlusion.
with varied
conditions.
An advance K.Oublal & 2021 MECS Normalizing Integrating Computational
Combination X.Dal Flow with the Normalizing Flow complexity due
of Normalizing YOLO into the YOLO to the
Flow and his innovative framework to integration of
YOLO to approach aims improve feature Normalizing
Detect Number to improve extraction and Flow, which
Plate robustness and representation for may affect real-
efficiency in number plate time processing
recognizing detection. This capabilities.
license plates includes training
under varying the combined
conditions. model on a diverse
dataset
An Automatic S.Kaur 2020 MECS Develop an Series of image The limitations
Number Plate efficient processing steps, of the paper
Recognition automatic including pre- include
System Under number plate processing, edge sensitivity to

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 445
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

Image recognition detection, and variations in

Processing (ANPR) system segmentation to lighting, angle,
leveraging isolate the number and occlusion,
image plate in recognition which can
processing techniques affect the
techniques. accuracy of
license plate
detection and
recognition.
ANPR Using I.Pustokhina, 2020 IEEE Approach aims Pre-processing The limitations
Optimal K- D.Pustokhina, Xplore to improve images to enhance of the paper
Means with D.Gupta accuracy and quality, followed include
CNN for efficiency in by the application potential
Transportation recognizing of optimal K-means challenges in
System license plates clustering for handling
within effective variations in
transportation segmentation of license plate
systems. license plate formats and
regions. environmental
conditions,
which may
affect
recognition
accuracy
Research on Shan luo And 2022 IEEE The objective of The methodology Limitations
Car License Jihong Xplore this research is involves using an include
Plate Liu to enhance car improved potential
Recognition license plate YOLOv5m for challenges with
Based on recognition precise license plate varying plate
Improved accuracy and detection followed designs,
YOLOv5m and efficiency by by LPRNET for environmental
LPRNET Paper integrating an accurate character conditions, and
improved recognition, the need for
YOLOv5m incorporating extensive
model for techniques like labelled datasets
detection with data augmentation for training.
LPRNET for and transfer
character learning.
recognition.
License Plate H. shi and 2023 IEEE The objective of Methodologically, limitations
Recognition D.zhao Xplore this paper is to it employs include

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 445
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

Based on enhance license advanced data susceptibility to

Improved plate preprocessing and varied lighting
YOLOv5 and recognition model training conditions and
GRU accuracy by techniques the need for
combining an large, diverse
improved datasets for
YOLOv5 model optimal
for detection performance
with a Gated
Recurrent Unit
(GRU) for
sequential
character
recognition
Multinational Chris 2020 IEEE Develop a The methodology limitations arise
License Plate Henry,Sung Xplore robust system includes the from varying
Recognition Yoon forrecognizing integration of license plate
Using Ahin,Sang- multinational diverse datasets and formats and
Generalized Woong Lee license plates by advanced machine potential
Character employing learning algorithms challenges in
Sequence generalized real-time
Detection character processing
sequence across different
detection countries.
techniques
Robust Ming-xiang 2020 IEEE The objective of The methodology limitations
Automatic Peng haq Xplore this paper is to includes a multi- involve
Recognition of achieve robust stage approach for handling
Chinese automatic detection and complex
License Plates recognition of recognition, backgrounds
in Natural Chinese license and variations
Scenes plates in natural in lighting and
scenes using angles that can
advanced image affect accuracy.
processing and
deep learning
technique
Automated Jithmi 2020 IEEE The objective of The methodology limitations
License Plate Shashirangana Xplore this survey involves analysing include the
Recognition: A Heshan paper is to and categorizing rapidly
Survey on Padmasiri provide a existing approaches evolving

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 446
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

Methods And Dulani comprehensive based on their technology

Techniques Meedeniya overview of effectiveness and landscape and
Charith various methods application context challenges in
Perera and techniques comparing
used in performance
automated metrics across
license plate different studies
recognition
(ALPR) systems.

IV. ACKNOWLEDGEMENTS changes caused by urban development and vehicle

growth. Innovation will be key.
We would like to thank everyone who contributed to The integration of ANPR with new technologies such
the Automatic Number Plate Recognition (ANPR) as the Internet of Things (IoT) and smart city projects
research project. First of all, our research Prof. holds great promise for creating transportation
A.V.Mote We thank them for their valuable guidance, networks. As machine learning continues to advance,
support and encouragement throughout this project. future license plate systems will become more
Their skills and understanding were essential to the accurate and flexible, making them relevant to
success of this work. modern traffic management and traffic lights. Finally,
ANPR technology has enormous potential to improve
V. CONCLUSION urban life, and continued investment in research and
development is critical to realizing this potential.
This study provides a comprehensive analysis of
automatic number recognition (ANPR), leading to VI. REFERENCES
several important findings. First of all, the integration
of image processing technology and machine learning [1]. M.K.Davy, P. J. Banda, and A. Hamweendo,
algorithms has increased the accuracy and efficiency ‘‘Automatic vehicle number plate recognition
of license plate recognition. Research shows that system,’’ Phys. Astron. Int. J., vol. 7, no. 1, pp.
using deep learning models for character recognition 69–72, Mar. 2023.
can produce great results, especially in complex [2]. A. V. Burkpalli, A. Joshi, A. B. Warad, and A.
situations such as changing lighting and occultations. Patil, ‘‘Automatic number plate recognition using
Pre-exercise strength is essential for improving full- TensorFlow and easyocr,’’ Int. Res. J. Mod. Eng.
body functions. The results of this research are quite Technol. Sci., vol. 4, no. 9, pp. 493–501, 2022.
significant. ANPR Advances in technology can lead to [3]. S. Rafique, S. Gul, K. Jan, and G. M. Khan,
advanced control solutions, improved control ‘‘Optimized real-time parking management
capabilities, and more efficient call distribution As framework using deep learning,’’ Exp. Syst.
cities grow, the ability to recognize vehicles could Appl., vol. 220, Jun. 2023, Art. no. 119686.
help improve traffic flow and create a safer city. The [4]. M. Jin, C. Sun, and Y. Hu, ‘‘An intelligent traffic
findings also highlight the need to continue to detection approach for vehicles on highway using
improve and update ANPR systems to meet the pattern recognition and deep learning,’’ Soft

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 447
 Swanand Joshi et al Int J Sci Res Sci & Technol. September-October -2024, 11 (5) : 439-448

Compute., vol. 27, no. 8, pp. 5041–5052, Apr.

2023.
[5]. M. Sankaranarayanan, C. Mala, and S. Mathew,
‘‘Efficient vehicle detection for traffic video-
based intelligent transportation systems
applications using recurrent architecture,’’
Multimedia Tools Appl., vol. 82, no. 25, pp.
39015–39033, Oct. 2023.
[6]. T. Azfar, J. Li, H. Yu, R. L. Cheu, Y. Lv, and R.
Ke, ‘‘Deep learning-based computer vision
methods for complex traffic environments
perception: A review,’’ Data Sci. for Transp., vol.
6, no. 1, p. 1, Apr. 2024.
[7]. V. Rajyalakshmi and K. Lakshmanna, ‘‘Detection
of car parking space by using hybrid deep Dense
Net optimization algorithm,’’Int.J.Netw.Manag.,
vol. 34, no. 1, p. e2228, Jan. 2024.

International Journal of Scientific Research in Science and Technology (www.ijsrst.com) | Volume 11 | Issue 5 448