You are on page 1of 6

2020 International Conference on Computer, Control, Electrical, and Electronics Engineering (ICCCEEE)

Vehicles Number Plate Recognition Systems


2020 International Conference on Computer, Control, Electrical, and Electronics Engineering (ICCCEEE) | 978-1-7281-9111-9/20/$31.00 ©2021 IEEE | DOI: 10.1109/ICCCEEE49695.2021.9429605

A Systematic Review
Mohammed Al Awaimri1, Sallam Fageeri2, Aiman Moyaid3, Abdullah ALhasanat4
'A^Department of Information Systems, College of EMIS, University of Nizwa, Sultanate of Oman
^Department of Electrical and Computer Engineering, College of EA, University of Nizwa, Sultanate of Oman
1{ 21482026 @uofn.edu.om},2{sallam,3aiman.moyaid, 4a.ismail @unizwa.edu.om}

Abstract- Automatic number plate recognition system (or


ANPR) is a system that uses optical character recognition to Fewer requirements. To this end, the automatic
read characters from solid images automatically and identification and recognition system of vehicles’ plate
immediately and then convert them to ASCII characters numbers were developed to manage and monitor mobile
readable by machines. Such a system has been widely used to
vehicles on roads [1].
recognize vehicles plate by using several algorithms and
Vehicles should be identified by their plate number
methodologies, including optical character recognition,
numbers. These plate numbers are readable for humans but
convolutional or deep neural network, morphological
not for machines since it is only an image. Because of that,
operations, and edge detection. This study aims at
understanding and analyzing the concept of the vehicle it needs an automatic vehicle plate recognition system that
number plate recognition system, especially those systems has the capabilities to transfer data from the real
which don’t need any human resources to accomplish their environment into digital systems. Vehicle number plate
missions. For this purpose, this paper presents an analytical recognition is a type of technology used to process captured
and theoretical comparison between several previous works in images of vehicle plate numbers and collect information
this field. According to this study, there are different levels of
about these vehicles. Automatic vehicle number plate
the recognition process starting from collecting images by
recognition system has many applications for traffic
cameras, detecting the region of plate numbers, segmenting
management such as automatic parking vehicles, control
the characters individually, comparing each number with the
border crossing, airports, monitoring traffic in roads, radar-
stored database, and ending with detected the whole plate
number. The performance is evaluated based on different based speed and driver seatbelt control, and many other

factors such as accuracy, precision, and recall. applications [2].

II. Li t e r a t u r e r e v ie w
Keywords: N u m b e r Plate, recognition system, detection,
Using machine and deep learning techniques, artificial
segmentation
intelligence, and internet of things technology to overcome
I. In t r o d u c t io n the difficulties and challenges that faced people in daily life
Recently, vehicle monitoring has received great attention and especially in the transportation sector such as traffic
due to the rapid increase of the vehicle numbers and their congestion, parking difficulties, accidents, and safety that
applications. Monitoring this huge number of vehicles need smart solutions to decrease it. In the last years,
need a rapid electronic system that has a fast response and automated and intelligent electronic solutions have been

978-1-7281-9111-9/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 07,2021 at 04:00:36 UTC from IEEE Xplore. Restrictions apply.
used to solve traffic problems such as autonomous cars, Another study in [5] discusses the automatic extraction methods

seat belts, or mobile phone usage detection applications [3]. of Iraqi number plates from captured images by using two
different methods: morphological operation and edge detection.
Automatic vehicle num ber plate recognition is defined as a
The authors have used 60 vehicle images from different types and
technology that uses character recognition on images to
conditions. The average run time for the morphological operation
read vehicle plates and then provide accurate vehicle
is 3s. Although it was good to extract number plates from low-
identification data. It can be used by traffic security to
resolution images, it takes a long time. As shown by this paper,
check if a vehicle is registered or licensed. Based on the extraction percentage rate of morphological operation is 98%.
different previous studies, vehicle num ber plate recognition The average run time for the edge detection method is 2s. It has a
system (VNPR) is also known as automatic license plate good performance, fast response, and simple design, however, it is
recognition (ALPR), license plate recognition (LPR), car plate not sensitive to noise. The extraction percentage rate was reported
recognition (CPR), automatic vehicle identification (AVI), vehicle as 82%.
recognition identification (VRI) and car plate reader (CPR) [4]. In the literature given in [1], authors discussed algorithms of
Vehicle number plate recognition technology is very important in automatic vehicle number plate recognition to identify Sudanese
the transportation system. For example: allowing cars that have license plates. The proposed system in this study has five levels:
permissions to park or enter local places, controlling and image input, image preprocessing, vehicle plate detection, plate
managing traffic in private or public organization parking, helping segmentation, and character recognition. As a result of this study,
the policemen to know information about the owners and drivers using 33 images of Sudanese vehicle number plates, the accuracy
of vehicles immediately in roads and also this system can help to rate of detecting and locating plate numbers from vehicle images
identify stolen vehicles in roads[4]. was 96%, while the accuracy rate of character extraction was
Vehicles number plate recognition software is an important latest 90%. Unfortunately, authors have faced the challenge that
developed technology that is used nowadays in transportation Sudanese plate numbers have a difficult format which is the
management. It is done in Malaysia, India, the United States, reason behind the accuracy rate of characters’ extraction.
United Kingdom, and other European countries. Also, there is
some practical implementation of using real number plates Research paper [6] discusses the effects of image distortion and

recognition system exist in some Arab countries like Saudi distance in character recognition. Specifically, the paper studies

Arabia, Iraq, and Sudan as shown in table 1 below. In what how the distance between the camera and car plate will affect the

follows, state-of-art algorithms and methodologies of the VNPR accuracy of character recognition. The proposed ANPR algorithm

will be discussed. Authors in [2] present ANN and object analysis in this study has three parts: plate number detection, segmentation,

algorithms to detect license plates in Saudi Arabia. According to and number recognition. The authors used zonal density as a

this paper, detecting license plates is achieved by combing features extraction method with a support vector machine

measured edges, morphological operations, and ANN. classifier to detect vehicle number plates. Support vector machine

Specifically, the proposed algorithm in this study has four phases: is a machine learning technique based on a supervised learning

image preprocessing, edge detection and analysis, object analysis, class. Using a dataset of 21 images captured atl, 3, 5 meters’

and ANN classifier. Image preprocessing and morphological distances and within zonesof 16x16, 64x64, 128x128, and

operations are used to improve the image quality. Then, Discrete 256x256. The reported accuracy is high at low distances and low

Wavelength Transform (DWT) is employed to decide whether the zonal density.

object is a license plate or not. Finally, an ANN classifier is


conducted to improve the detection rate. The input to ANN is the The proposed technique of identifying vehicle plate numbers in

output from DWT. Results of this study proved that horizontal literature [7] is built on morphological operations. This system

coefficients, demy family, and the trains training algorithm get the used optical character recognition device to read characters from

best detection rate of 99%. vehicle images. The proposed system is divided into different

Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 07,2021 at 04:00:36 UTC from IEEE Xplore. Restrictions apply.
levels: image capture, image preprocessing, character extraction, percentage rate, that is more than 95%, is obtained. CNN model
character segmentation, character matching, number plate presents better results than artificial neural networks and
identification, and the final output. The output is twofold: license probabilistic neural networks.
plate number location and recognition with template matching. The authors of the research paper [9] discussed optical character
The proposed algorithm is evaluated using 150 vehicle images recognition (oCR) for number plate recognition using a neural
captured at 4 to 5 meters’ distance. The performance rate of the network. This system is divided into three phases: number plate
number recognition is about 92% and the computation time was localization, character segmentation, and optical character
about 0.3 seconds. Moreover, the proposed algorithm can identify recognition matching. Using 300 vehicle images from different
the characters up to 45 degrees while the performance rate of the countries, the accuracy rate of the proposed localization scheme is
conational algorithm is 89%. Despite this, cases such as abnormal 96.7% and the accuracy rate of character recognition is 92.2%.
plate number sizes and capturing the image at large distances have overall, the whole automatic number plate recognition has
caused this algorithm unsuccessful. reported an accuracy rate of about 94.45%.
Authors in [10] present two methods of vehicle plate recognition
Literature study in [8] presents a deep learning algorithm for systems in Libya: region-based features and probabilistic neural
automatic number plate detection and recognition. The proposed network method. The methodology of this system is divided into
system in this study has three main levels: detection, five stages: image collection and preprocessing, license plate
segmentation, and character recognition. The convolutional neural detection, character segmentation, feature extraction, and number
network for classifying the plates from non-plates is employed. plate recognition. The recognition is done by using a template
Two datasets have been collected to test and evaluate the matching and probabilistic neural network (PNN) classification.
performance and accuracy of the system. Each dataset consists of The dataset considered by the authors consists of 100 Libyan
around 2400 images captured in different weather conditions and vehicle images which have been used to evaluate the accuracy of
is divided into four main sections. The proposed number plate the proposed automatic recognition system. The performance of
detection system is evaluated in terms of precision, recall, and f- this system is measured by three metrics: precision, recall, and f-
measure. As a result, a high accuracy rate for plate detection is score and are compared to another existing study done in Libya.
attained in some cases, while it fails in some other cases, The performance measurements based on PNN are 84% recall,
particularly under extreme lighting conditions (i.e., darker or more 81% precision, and 83% f-score. On the other hand, based on the
bright). The performance of this recognition system is measured feature classification method these performance measurements
based on the character recognition accuracy, where a high are83% recall, 80% precision, and 81% f-score.

Ta bl e 1: Su m m a r y o f l i t e r a t u r e r e v ie w

Reference Title Materials and Methods Results


Saudi Arabia license Proposed methodology has four stages: image Results of this study prove that horizontal
plate detection based on preprocessing, edge detection and analysis, coefficients, Demy family and the trains training
ANN and objects object analysis and (ANN)Artificial neural algorithm get the best detection rate 99%.
analysis [2] network classifier.
Automatic Iraqi cars Using two methods: morphological operation and Average run time for morphological operation 3s and
number plates extraction edge detection to extract Iraqi car number plates it is good to extract from low resolution images but it
[5] from images. takes longer time. Extraction percentage 98%.
Using MATLAB R2014a to build the system and Average run time is 2s for the edge detection
60 car images. method. It has a good performance, fast and simple
but it isn’t sensitive to noise. Extraction percentage
82%.

Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 07,2021 at 04:00:36 UTC from IEEE Xplore. Restrictions apply.
Sudanese license plate Using 33 images of Sudanese vehicle number 96% of images are detecting and locating plate
identification using plates. The proposed system in this study have number and 90% of images are successful in
automatic number plate five levels: image collection, image character recognition but authors have faced a
recognition [1] preprocessing, vehicle plate detection, plate challenge which is Sudanese number plates.
segmentation and character recognition.
Long distance automatic 21 images of vehicles plates from 1, 3, 5 meters’ The accuracy of recognition is high in 1-meter
number plate distance & (SVM) support vector machine with distance and decrease in 3- and 5-meters distance
recognition under zonal density as a classifier. ANPR algorithm
perspective distortion have three parts: detection of plate number,
using zonal density and character segmentation and number recognition
support vector machine
[6]
Automatic vehicle 150 images collected from different locations. Recognition rate is 92% and computation time is
number plate This system is built into morphological about 0.3 seconds and this is able to identify the
recognition using operations. characters up to 45degrees
structured elements [7]
Deep learning system This research using deep learning approach Number plate detection system is evaluated in three
for automatic license (convolution neural network model for plate terms: precision, recall and f-score rate.
plat detection and classification) for number plate detection and Recognition performance of the system is measured
recognition [8] recognition. Three main levels are in this system: by character recognition accuracy rate which is
detection, segmentation and character higher than 95%.
recognition. Two datasets of images from
different situations.
Improved OCR based Using 300 vehicle images from different The accuracy rate of localization is 96.7% and
automatic vehicle countries. Artificial neural network based optical accuracy rate of character recognition is 92.2% and
number plate character recognition algorithm used in ANPR the complete automatic number plate recognition rate
recognition using system. is 94.45%.
features trained neural
network [9]

Libyan vehicle plate 100 Libyan vehicle images & two methods of The performance of this system is evaluated by three
recognition using region extraction and recognition: region-based features metrics: precision, recall and f-score and it is
based features and and probabilistic neural network method. compared to another existing study done in Libya
probabilistic neural
network[10]

III. Re s e a r c h met h od

According to different research papers and journal articles


presented in the literature, an automatic plate number recognition
system has four basic steps: input images and preprocessing,
vehicle plate detection and localization, character segmentation,
and character recognition as shown in figure1.

Fig. 1. Vehicle plate number recognition system[11]

A. Image collection and preprocessing: using digital


cameras to capturing images of vehicles. The image must
have a plate number. The captured image is affected by many

Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 07,2021 at 04:00:36 UTC from IEEE Xplore. Restrictions apply.
factors and it needs a preprocessing stage to correct the issues
and errors in images. Zonal density with support vector machine classifia'

B. Vehicles plate detection : detecting the rectangular shape of


vehicle plate from the taken images.

C. Character segmentation: it is a technique of dividing


images into multiple sub-images of symbols or characters.
Dividing the plate number into different parts or objects.
D. Character recognition: it is a process of identifying
Capturing distance
characters from the input image and converting it to
meaningful text. Portioned characters will be compared with Fig. 3. The relationship between capturing distance and
the stored template to find the best match. recognition rate
IV. R ESULTS
This part of the research explains the results of comparing
different recognition algorithms as discussed before. The selected
Figure 3 describes the effects of capturing distance between the
algorithms in this comparison include Artificial Neural Networks
camera and vehicle plates on the recognition process. As shown in
(ANN), Morphological operation, Edge detection, and
this figure, an inverse relationship between distance and plate
Convolution NN.
recognition rate is noticed. The closer distance between the

■ Recognition rate % ■ Plate region detection or extraction % camera and plate number increases the rate of recognition.

Measure the recognition rate %

Fig. 2. Recognition rate of different algorithms

Figure 2 explains the recognition rate of the selected algorithms.


The recognition rate is ranged between 99% and 82% of the
sample that used to check the accuracy of each recognition
Using three metrics precision, recall and f-score to calculate the
algorithm. The best accuracy rate is reported by the Artificial
false and true recognized plates based on two performance
Neural Networks (ANN), while the worst case is noticed when the
measurement probabilistic neural network (PNN) and region-
Edge detection method is used. Also, using the morphological
based features (RBF) is shown in Figure 4. As a result, the
operation to convert solid plate images to readable machine
accuracy rate of PNN is greater than RBF. The precision rate is
characters has demonstrated a high accuracy rate that is 98%.
less than recall and f-score in both methodologies.

Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 07,2021 at 04:00:36 UTC from IEEE Xplore. Restrictions apply.
Based on our discussion in the above results, we have defined the recognition has shown to be is the most popular and useful choice
advantages and challenges of using an automatic vehicle plate because it takes less time and results in high recognition accuracy
number recognition system as follows rate. The discussed studies in this paper are based on real because
they are using real plate numbers to evaluate the performance of
A. Advantages each algorithm in several situations
There are a lot of advantages for individuals and organizations
when developing an automatic number plate recognition system. REFERENCES
For example, it can help in parking vehicles, where a fast checking [1] H. MUBARAK, A. O. IBRAHIM, A. ELWASILA, S.
about empty parking is achieved and time-wasting is avoided. BUSHRA, I. TECHNOLOGY, AND K. NORTH, “SUDANESE
Allowing authorized vehicles to enter local locations such as LICENSE PLATE IDENTIFICATION USING AUTOMATIC
employees of one company without adding another tool in NUMBER PLATE RECOGNITION.”
vehicles. Increasing the security in local areas and prevention of [2] A. M. Abbas and A. E. Rashid, “Saudi Arabia License Plate
car theft. Representing statistical reports about vehicle traffic Detection based on ANN and Objects Analysis,” vol. 12, no. 13,
inroads that help to manage the traffic and checking if there are pp. 3740-3751, 2017.
unauthorized or stolen vehicles. Also, it can be used in smart [3] H. Saghaei, “Proposal for Automatic License and Number
control of gates and traffic lights [3], [4]. Plate Recognition System for Vehicle Identification,” no. October,
2016.
B. Challenges [4] S. Ghasempour, “Automatic License Plate Recognition
Detecting and extracting characters from digital images may have (ALPR),” 2015.
some challenges. For example, darkness or shadow and bad [5] S. S. Omran and J. A. Jarallah, “Automatic Iraqi Cars Number
weather like rain or fog can negatively affect the performance of Plates Extraction,” no. 1.
the vehicle plate number recognition system. This is due to bad [6] S. Wibirama and H. A. Nugroho, “Long Distance Automatic
quality of images captured in such conditions and it may break Number Plate Recognition under Perspective Distortion using
down the security and control system. Also, the distance and Zonal Density and Support Vector Machine,” pp. 1-6, 2017.
rotation angle can largely affect the quality of vehicle images and [7] R. Islam, K. F. Sharif, and S. Biswas, “Automatic Vehicle
in turn, reduce the recognition accuracy rate. Next, privacy Number Plate Recognition Using Structured Elements,” 2015
concerns play an important role in improving the quality of the IEEE Conf. Syst. Process Control, no. December, pp. 44-48,
recognition system. For example, people are afraid about their 2015, doi: 10.1109/SPC.2015.7473557.
information if someone else stole it and use it badly. Furthermore, [8] Z. Selmi et al., “Deep Learning System for Automatic License
the most common challenge is the difference between vehicle Plate Detection and Recognition,” no. November, 2017, doi:
number plates with respect to its format and style, colors and 10.1109/ICDAR.2017.187.
combination of Arabic and English letters and numbers [2], [10]. [9] B. V Kakani, D. Gandhi, and S. Jani, “Improved OCR based
Automatic Vehicle Number Plate Recognition using Features
V. Co n c l u s io n
Trained Neural Network,” 2017.
This paper presented a summary of vehicle number plate
[10] K. A. Jabar and M. F. Nasrudin, “Libyan vehicle plate
recognition systems used to identify vehicles from their plate
recognition using region-based features and probabilistic neural
number. A vehicle number plate recognition system is developed
network,” J. Theor. Appl. Inf. Technol., vol. 94, no. 1, pp. 104-
using different image processing algorithms. Although the
114, 2016.
processing levels are mostly the same in different studies starting
[11] S. Kranthi, K. Pranathi, and A. Srisaila, “Automatic Number
from collecting and preprocessing plate images until character and
Plate Recognition,” vol. 2, no. 3, pp. 408-422, 201
number recognition, they have provided different accuracy rates.
Using deep learning algorithms such as neural networks in

Authorized licensed use limited to: University of Prince Edward Island. Downloaded on June 07,2021 at 04:00:36 UTC from IEEE Xplore. Restrictions apply.

You might also like