0% found this document useful (0 votes)
201 views5 pages

Car Number Plate Detection

This document summarizes a research paper on automatic vehicle number plate detection using deep learning techniques. It discusses how number plate recognition (NPR) systems can be used for applications like access control, parking management, toll collection and traffic monitoring. It then describes the key steps in the proposed methodology - number plate detection using image segmentation techniques followed by character segmentation and recognition using a Convolutional Neural Network model. The experimental results showed an accuracy of 94.94% for the character recognition step.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
201 views5 pages

Car Number Plate Detection

This document summarizes a research paper on automatic vehicle number plate detection using deep learning techniques. It discusses how number plate recognition (NPR) systems can be used for applications like access control, parking management, toll collection and traffic monitoring. It then describes the key steps in the proposed methodology - number plate detection using image segmentation techniques followed by character segmentation and recognition using a Convolutional Neural Network model. The experimental results showed an accuracy of 94.94% for the character recognition step.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Car Number Plate Detection


Sharon Babu K.1, Dr. K. Arthi2
Student1, Associate Professor2
Dept. of Data Science and Business
StudiesSRM Institute of Science and Technology
Kattankulathur, India

Abstract:- Vehicle tracking is no longer feasible due to II. SCOPE


population growth. It's a waste of time and resources.
Tracking individual vehicles has become a very difficult NPR can be used in housing societies or apartments to
task due to the tremendous development in the vehicular only letthe resident’s vehicle inside by storing the number
sector on a daily basis. plate de- tails in a database and cross checking it. This can
reduce the man force near the gate. NPR can also be used in
This work proposes an automatic vehicle parking ar- eas to track down the vehicles violating the
monitoring sys- tem for fast-moving automobiles using rules, ticketless parking, parking automation, parking fee
roadside sur- veillance cameras. In modern smart cities, charging, fraud by changing tickets are some areas where
license plate recognition systems are used in toll NPR is useful. NPR can be also extended to motorway
payment systems, parking charge payment systems, and road tolling where efficient road tolling helps to reduce
residential access control. These electronic technologies scam linked to non- payment, makes charges effectives and
are not only useful in people's daily lives, but it also hence thereby decreases work- force to course events of
provides management with safe and efficient services. exclusions [1]. NPR can also be used in journey time
An effective approach for rec- ognizing Indian measurement data where the data collected by NPR and
automobile number plates has been im- plemented in the feeding back information to the drivers to increase the
suggested algorithm. security at traffic, optimizing traffic routes and reducing cost
and time.
The proposed system is able to deal with noisy, low
illumi- nated, cross angled, non-standard font number III. NUMBER PLATE DETECTION
plates. This thesis presents effective deep learning-based
ALPR (Au- tomatic License Plate Recognition) model The factors that are to be considered for numberplate
using Character segmentation and CNN (Convolutional detectionare:
Neural Network) based recognition model. The  Size of the plate: Different vehicles will be having
experimental result gives an accuracy rate of f1 score of different plate sizes
94.94%.  Background color: Different vehicle types will be having
different number plate background, for exam- ple a taxi
I. INTRODUCTION car will be having yellow plate background whereas a
private vehicle will be having white plate background.
Vehicle NPR (Number Plate Recognition) or LPR  Location of the plate: Different plates will be in dif- ferent
(License Plate Recognition) is an increased CV (Computer locations of the vehicle.
Vision) technology that relates vehicle while not direct
human affilia- tion through their range plates. Every day, the We extract number plates with the help of image
amount of vehi- cles on the road is constant to grow. Due to segmenta- tion. Most common method is image
this, the news un- folds nearly on a daily basis regarding the binarization. Otsu’s method can also be used to convert
vehicle being robbed from the parking space or the other images into grayscale.
place inside town or having an accident and theft. To
acknowledge these, vehi- cles establishments ought to thus A. Image binarization
install variety plate detection and acknowledgement device Image binarization is used to convert an image into
on CCTV (Closed Circuit Tele- vision) at each corner in blackand white form. We choose a certain threshold value to
each region. This method improves the police’s aptitude to differ-entiate pixels some as black and some as white.
trace prohibited activities linking the use of automobiles. This is a bitdifficult but can be overcome using adaptive
NPR systems square measure effectively used by provincial thresholding.
institutions and producing teams all told aspects of safety,
examination, traffic supervision tenders.

IJISRT22JUN058 www.ijisrt.com 94
Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
OCR also known as Optical Character Extraction is a
charac- ter extraction tool which has numerous software
available for character extraction such as tesseract which is
an open-source OCR developed by Google. Tesseract has
been able to achievea rate of 98.7% for character extraction
[2].

ANN contains inter connected artificial neurons. There


are so many algorithms based on ANN. For classification of
charac- ters MLP ANN model is used. It has three layers,
input layer, hidden layer and output layer. The input layer is
for decision making whereas the hidden layer computes
Fig. 1: Image converted to grayscale complicated asso- ciations and finally the output layer for
the result. We train ANN using feed forward back
B. Edge detection propagation. ANN was able to achieve a rate of 99% in
Edge detection is employed for feature withdrawal character extraction [3].
and have choice. it's totally troublesome to use edge Template matching can be used for characters that are
detection in complicated pictures as there may well be fixed size. It's principally utilized in face detection and
ensuing with object boundaries with no connected curves medical im- age process. It has two components: template-
based matching and feature based matching. If the template
image has strong features, feature based matching can be
used. Template match- ing was able to achieve about
95.7%-character extraction rate [4].

Fig. 4: Character extraction

IV. PROPOSED METHODOLOGY

Fig. 2: Edge detection The rising population and technological advancements


have drawn attention to the area of transportation. The
C. Character segmentation amount of cars on the road has expanded rapidly, which is
After finding the location of the number plate, we then one of the primary causes of traffic congestion and
check the characters. There are many methods to perform violations. NPR was created to decrease traffic offences
character segmentation. We can apply the above discussed while also automating traf-fic management. In India,
image binarization to character segmentation. For Indian different NPR tactics are used, however their effectiveness
number plates, blob coloring is not suitable. We should is relatively low. The proposed model focuses on the
apply image scissoring algorithm where the plate will be detection of car number plate and recog- nizing it number
scanned vertically and when white pixel is not found it will plate. Various algorithms are implemented in python using
be scis- sored at that row and the values will be stored in a Russian blurring method, OpenCV and CNN[5].
matrix.
The first thing we have to do is data understanding. We
input the image of the plate and then our aim is to extract the
char- acter images. For that, we perform character
segmentation which is the most important step for
identifying the number plates.

In brief as the first step, we input the image then we


Fig. 3: Segmented characters
convert the image into grayscale or black and white. Then
D. Character extraction we perform image binarization proceeded by blurring the
Character extraction identifies and converts the image image to reduce the unwanted louds. Then we perform
text into actual text. All the number plate identification character segmentation to take out each and every character,
methods use character extraction and some of those methods then do the character extraction methods using OCR or
can be discussed here. Most common methods used in Template matching or ANN.
character extraction are OCR (Optical Character
Recognition), ANN (Artificial Neural Net- work) and
Template matching.

IJISRT22JUN058 www.ijisrt.com 95
Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
A. Understanding the Libraries character extraction using var- ious character extraction
Numpy, OpenCV, Tensorflow, SKlearn tools such as OCR, ANN, Template matching etc.

NumPy is a python package used for various


applications. Aside from the obvious scientific
applications, it also has a number of other uses. NumPy
may be used to store common data in a multi-dimensional
container.

The following are some of NumPy’s features:


 NumPy Array is a multidimensional array with vectorized
arithmetic that is quick and space-efficient.
 sophisticated broadcasting features and opera- tions
 Standard mathematical functions can be used to perform
quick operations on big sets of data.
 Without the use of loops, data may be collected.
 Working with memory-mapped files and read- ing/writing
array data to disc
 Toolkits for integrating C, C++, and Fortran pro- grams Fig. 5: Flowchart
with linear algebra, random number gen- eration, and other
capabilities. C. ANN
Open-Source Computer Vision is an open-source The ANN has mainly three layers: input layer, hidden
computer vision and machine learning software library. It is layerand output layer.
used to identify various objects, faces, identify action The input layer is for decision making whereas the
through photos or videos and has various specialties. hidden layer computes complicated associations and finally
TensorFlow is another Python library used in Deep the output layer for the result. We train ANN using feed
Learning models with the help of keras. It has various tools forward back propagation. ANN was able to achieve a rate
like Image Data Generation, Dense, Flatten, MaxPool and of 99% in char- acter extraction [6].
more. Each input will have its own weights and after entering
Scikit-learn (Sklearn) provides various tools for input layer, it undergoes various preprocessing steps and at
modelling ML algorithms which includes classification, hidden layer there undergoes convolution and pooling and
regression, cluster- ing, SVM, PCA and dimensionality after reach-ing a certain threshold value, the processed
reduction and many more. These are implemented using segmented image is then transferred to output layer where
NumPy, SciPy, and Matplotlib libraries. we will be obtaining the results.

The following are some of Sklearn's key features:


 Clustering is a method for arranging unlabeled data into
groups.
 Dimensionality reduction is a method for reducingthe
number of characteristics in data so that it can be summed
up, visualized, and features chosen.
 As the name indicates, ensemble methods integrate the
predictions of several supervised models.
 Feature extraction is a method of extracting data
characteristics and defining attributes from images and
text.
 Feature selection is a method for discovering desir- able
characteristics that may be used to create super- vised Fig. 6: ANN layers
models.
D. CNN
B. Flowchart CNN comprises of various layers of artificial neurons.
After inputting the image, it is then converted to gray Thefirst layers detect the edges such as diagonal, vertical
scaleand then undergo various stages of filtering. Here and hor- izontal edges. The second layer detect the more
filtering refers to blur the image to reduce the unwanted complex fea- tures like edges and corners. The deeper layers
edges and louds from the image so we use either Median detect highlevel features like face, object etc.
blur or Gauss- ian blur. These steps are considered as image
pre-processing steps. After filtering, then we perform edge The vehicle imaged after input undergoes image pre-
detection methods, here we use canny edge detection. After process- ing that is converting into gray scale and applying
detecting the plate, we then go for character segmentation blur to re- duce the loud. Then it undergoes edge detection,
which is considered as an important step as it is necessary to we use canny edge detection here. After all the filtering
identify each and every character image. Then we form process, it under- goes horizontal detection, then vertical

IJISRT22JUN058 www.ijisrt.com 96
Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
detection and finally we apply character segmentation to The objective of this project was to highlight that free
identify the character im- ages and then apply character and open-source technologies are competent enough to be
extraction to extract the vehicle number plate. used in machine learning fields. The device performs
admirably under a wide range of lighting situations and with
various types of number plates usually seen in India.
Even though there are acknowledged limitations, it is
unquestionably superior than existing proprietary solutions.
This was developed in Jupiter notebook using python
language by taking images as input from the system. After
80 epochs it gives a f1 score accuracy of 94.94% [7].

VI. FUTURE ENHANCEMENTS


The automatic vehicle identification system will play a
Fig. 7: CNN
signif- icant role in identifying threats to defense in the
E. KNN (K-Nearest Neighbours) future. It can also boost women's security since they can
A KNN machine learning structure for programmed readily recognize the number plate before utilizing cabs or
auto- mobile authorization number plate extraction was other services. If a bright and crisp camera is employed, the
matured. The KNN classifier is going on secondhand system's resilience can be increased. The government should
in accordance with the sides of the license to recapture the take an interest in improving this technology because it is
enrollment plate from the p h o t o . They need associate both cost-effective and environmentally benign when used
degreed gone through had undergone passed through efficiently in a variety of settings.
responded suffered that an au- thorization code is
Sensors at the entry can be used to improve the system.
correct labelled through their submittedkind and earned an
When a car enters the gate, the image of its license plate is
accuracyon top of ninetieth. Image extrac- tion algorithms
captured. The image is then transferred to MATLAB and
in moving neural networks will identify every- thing, from
subsequentlyto a database, where it is classified. If the
text to ideas, visual and audio recreation transmit- ted via
license plate match- es, the gate will open; otherwise, an
radio waves files, and videos. Neural networks are a mesh
alarm will besent.
set of neurons or perceptron’s named growth. every bud
uses a alone recommendation information, principally one The precision of the recognition is obviously the most
component of the countenance, And uses a lucid signifi- cant aspect of this technology. As a result, in order
estimation named an incitement perform that produces to over- come the accuracy constraints, this program needs
results and every somatic cell incorporates a mathematical be im- proved and updated. To improve the precision of the
score that decides attract impact. A high-routine-based recogni- tion, we need include certain preprocessing steps to
system for automobile range plate extraction has been eliminate interferences. Furthermore, we would continue our
matured. They need used edge-based countenance prepare research into license plate identification in more difficult
strategies for authorization plate discovery and have too settings, such as cars at night or in severe weather, and so
used a three-layer feedforward faux interconnected system on. If we were to achieve all of our goals, our application
for vehicle range plate integrity ac- knowledgment utilizing would have a very bright future.
a learning invention for back-propa- gation.
ACKNOWLEDGMENT
V. CONCLUSION
I express my humble gratitude to
NPR programmed number plate applications are Dr.C.Muthamizhchelvan, Vice-Chancellor, SRM Institute
strongly in- creasing in Indian platform since there is of Science and Technology, for the facilities extended for
phenomenal increase in no of cars, bike, auto and Industry the project work and his contin- ued support.
vehicles. NPR can be used in various applications including
toll collection, Mall parking spaces, even manage vehicle in I extend my sincere thanks to Dean-CET, SRM
parking spaces, and it can also be included in traffic Institute of Science and Technology, Dr.T.V.Gopal, for his
cameras which can be used to monitor and regulate traffic invaluable support.
violators, etc., This project has many benefits. Character
segmentation was used to extract every number from the I wish to thank Dr.Revathi Venkataraman, Professor
number plates. Finally, F1 score preci- sion, recall was & Chairperson, School of Computing, SRM Institute of
found for the input image and the result was calculated. The Science and Technology, for her support throughout the
project has its own limitations as the model is not using project work.
higher machine learning or deep learning algorithmsbut it
works efficiently for an average use case and that will I am incredibly grateful to our Head of the
betaken as a future work in scaling up the project to the Department, Dr.M. Lakshmi, Professor, Department of
further level. Data Science and Busi- ness Systems, SRM Institute of
Science and Technology, for her suggestions and
encouragement at all the stages of the project work.

IJISRT22JUN058 www.ijisrt.com 97
Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
My inexpressible respect and thanks to my guide, Dr.
K. Arthi, Associate Professor, SRM Institute of Science and
Technology, for providing me with an opportunity to pursue
my project under her mentorship. She provided me with the
freedom and support to explore the research topics of my in-
terest. Her passion for solving problems and making a
differ- ence in the world has always been inspiring.

I sincerely thank the Data Science and Business


Systems staff and students, SRM Institute of Science and
Technology, for their help during our project. Finally, we
would like to thank parents, family members, and friends for
their unconditional love, constant support, and
encouragement.

REFERENCES

[1.]Abhay Singh, Anand Kumar Gupta, Anmol Singh, Anuj


Gupta, Sherish Johri, “VEHICLE NUMBER PLATE
DETECTION USING IMAGE PROCESS- ING”,
Department of IT, Volume: 05 Issue: 03 | Mar-2018
[2.]Ganesh R. Jadhav, Kailash J. Karande, “Automatic
Vehicle Number Plate Recognition for Vehicle Park- ing
Management System”, IISTE, Vol.5, No.11, 2014.
[3.]Mutua Simon Mandi, Bernard Shibwabo, Kaibiru Mutua
Raphael,” An Automatic Number Plate Recognition
System for Car Park Management”, In- ternational
Journal of Computer Applications, Vol- ume 175 – No.7,
October 2017

[4.]M.T. Shahed, M.R.I. Udoy, B. Saha, A.I. Khan, S.


Subrina, “Automatic Bengali number plate reader,”
IEEE Region 10 Annual International Conference,
Proceedings/TENCON, 2017-Decem, 1364–1368, 2017,
doi:10.1109/TENCON.2017.8228070.
[5.]Regtransfers.co.uk, “Number plates rules”. [Online].
Available: https://www.regtransfers.co.uk/acrylic-
number-plates/number-plates-rules
[6.]R. Islam, M.R. Islam, K.H. Talukder, “An efficient
method for extraction and recognition of bangla char-
acters from vehicle license plates,” Multimedia Tools
and Applications, 79(27–28), 20107–20132, 2020,
doi:10.1007/s11042-020-08629-8.
[7.]S. Saha, “A Review on Automatic License Plate
Recognition System,” ArXiv, 2019.

IJISRT22JUN058 www.ijisrt.com 98

You might also like