Two Wheel Vehicle License Plate Detection System with Image

Processing

Abdullah Sani and Dhea Ade Rahmadani

Department of Electrical Engineering, Politeknik Negeri Batam, Batam, Indonesia

Keywords: Image Processing, Vehicle License Plate Detection, K-Nearest Neighbor

Abstract: Image processing is a method used to process images into a digital form, which serves to improve and improve

the quality of an image. One of its utilization is to detect the license plate of the vehicle that enters the campus

area. Currently for the identification of vehicle license plates in the parking area of Politeknik Negeri Batam

Campus still use the manual system that is, when the vehicle enters the parking area then the parking attendant

only gives the card. Thus, there is no data storage of vehicles coming into the parking area such as vehicle

license plate recording. With the license plate detection system using image processing techniques can

facilitate the process of checking the license plate of incoming vehicles by mounting the camera to check the

license plate number, as well as using the method K-Nearest Neighbor to recognize each character that is on

the license plate automatically, this method has an accuracy level of about 84.5% in 80 characters.

1 INTRODUCTION

In this modern era, the development of science and

technology that is progressing in all areas has a

positive impact on people's lives. With advances in

the field of technology, more and more new

technologies are emerging. One of the technologies

that develop in the field of digital image processing.

Image processing is a method for the processing of

images (Image) into digital form for a specific

purpose that serves to improve and improve the

quality of an image to provide information in the form

of objects detected on an image (Triyandil, 2014).

One of its applications is to identify the license plate

number that is entered in an institution such as a

parking area campus near a postal park. In general, to

recognize the license plate number that enters the

campus parking area still use the manual system that

is, when the vehicle enters the parking area, the

parking attendant gives the card and there is no data

storage of incoming vehicles such as vehicle license

plate registration. The way is ineffective and will

complicate the process of identification of vehicles

entering the parking area. The manual parking system

will consume a lot of time and energy (Nur Taufiq,

2012).

Meanwhile, Santra et al. (2019) implemented

Signal to noise ratio, PSNER and mean square error

to computed each capture image after detection. The

objective of implementation is to primarily detect

moving vehicle. Others researcher Muhammad Tahir

Qadri, Muhammad Asif et al. (Qadri, 2009)

implemented Automatic Number Plate Recognition

(ANPR) as an image processing system to detect the

vehicle number (license) plate to identify the vehicle.

The system captures the vehicle image and then

extraction them using image segmentation. By using

optical recognition technique to recognize each

character in the license plate.

K Mahesh Babu, M V Raghunadh et al. (Babu,

2016) implemented the template matching method to

recognize the characters in vehicle license plate. The

researcher conducted four major steps to detect the

character as follows: preprocessing of capture image,

extracting license number plate region, segmentation

and characters recognition.

In this paper, made a system that facilitates the

process of checking the license plate of the incoming

vehicle, by mounting the camera to detect the license

plate of the vehicle. The detection process begins with

data retrieval with the camera mounted (Budianto,

2015). This vehicle license plate check, able to

recognize the character of letters and numbers found

on the license plate that enters the parking area using

the image processing method and K-Nearest

Neighbor which can be useful as a verification of the

Sani, A. and Rahmadani, D.

Two Wheel Vehicle License Plate Detection System with Image Processing.

DOI: 10.5220/0010351900750080

In Proceedings of the 3rd International Conference on Applied Engineering (ICAE 2020), pages 75-80

ISBN: 978-989-758-520-3

identity of the vehicle number when entering the

parking area of the campus more efficient and simple.

2 BASIC UNDERSTANDING

2.1 Digital Image Processing

Digital image processing is a process of processing

and analysis of two-dimensional image imagery that

uses computers, aiming to improve the image quality

for easy interpretation by humans or machines

(computers) (Sari, 2014). A digital image can be

represented by a matrix of two dimensions f (x, Y)

consisting of M for columns and N for rows, which

between the column and row intersection is also

called the pixel or the smallest element of an image.

There are several types of digital imagery that are

often used namely, RGB image, binary image, and

grayscale image.

RGB Image

RGB image is a color image which consists of three

basic colors, namely Red (R), Green (G), and Blue

(B). Each basic color has its own intensity with a

value range from 0 to 255 (8 bits) (Sari, 2014) as

shown in Figure 1 is an example of an RGB image on

a vehicle plate.

Gray Image Concept

Gray imagery is a digital image processing that takes

the value of the different degrees of the sensitivity of

each pixel. Where a pixel is represented by a range of

values from 0 to 255, a value of 0 represents for the

minimum intensity or image condition in a dark state

and the 255 value represents for maximum or light

intensity. We can see in Figure 2 is the result of the

gray image.

Figure 1: RGB image.

Figure 2: Gray image.

Image Binarization Concepts

Image Binarization is a process that aims to change

the intensity value of an image to 2 values of 0 and 1.

Each pixel consists of only a black color of 0 and a

white color worth 1 (Kusumanto, 2011). To do this

process used threshold can be adjusted value

according to the wishes (Setiawan, 2015). The image

binarization is often used in research studies using

segmentation and morphology processing processes.

Figure 3 is the result of an image binarization.

Canny Edge Detection

Canny edge detection is a filter process used to detect

edges of an image. Canny edge detection was

developed by John F. Canny in 1986. The method of

canny edge detection is used to look for a contour that

is considered a vehicle number plate. Figure 4 is the

result of an image binarization.

2.2 Connected Component

Connected components are a collection of pixels of

the same value, and that are interconnected with each

other through pixel connectivity, i.e. all pixels in the

connected component have the same pixel intensity

value, in some ways connected. Once all the groups

have been specified, each pixel is labeled with a gray

or color labeling according to the component set. This

method is used in the process of character detection on

vehicle number plates, in Connected Component

methods. The process of analysis of imagery objects

such as calculating the area, height, or width of the

object, from the candidate object to be carried out a

rule Filtration that represents a license plate.

Filtration can be done by doing calculations to

determine the area of an object that has been labeled

(Ruslianto, 2011).

Figure 3: Image binarization.

Figure 4: Canny edge detection.

ICAE 2020 - The International Conference on Applied Engineering

2.3 K-Nearest Neighbor

K-Nearest Neighbor is a method that uses a

supervised algorithm were the results of a new query

are classified based on the majority of categories on

the KNN. The function of KNN is to classify the

characters of letters and numbers found on the motor

plates. KNN sets the class label type on most K-

nearest patterns on the data train to be utilized, the

similarity of patterns in the data train must be defined

first (Tauchid, 2015). K is the number of closest

neighbors. The number of neighbors is a core

determining factor. The way the KNN works is if P1

is the core, then next performs a label prediction and

looks for the closest point K with P1 then classifies

the points with the highest number of neighbors in

that area. Each class has a value that represents its

respective class. Most values will be taken as

predictions. For example, K = 4, the closest point in

Class A, Class B, Class B, Class B so that the highest

value is in class B Then the prediction result of the

new data is Class B as shown in Figure 5.

The KNN on the OpenCV function uses a flatten

image to be a single row/one column array. Each

dataset will be training to see the pattern of each

character which will then be used as a comparison for

the character recognition of the trial as can be seen in

Figure 6.

Figure 5: KNN Concept.

3 PROPOSED METHOD

The object to be detected by the camera is a two-

wheeled vehicle license plate. Images that are capture

by the camera will be processed by the computer

using the K-Nearest Neighbor method, then the

detection result is displayed on the computer layer. A

more detailed process can be seen in Figure 7.

Figure 7 presents the way the overall system

works. The system itself is divided from 4 stages,

namely pre-processing image, plate detection,

character segmentation, character recognition. At pre-

After that, the process to detect the plate part of

the motor vehicle by using canny edge detection.

Then from an edge image that is already in the get

filtered out based on the form contours. The shape of

the contours considered a plate is a rectangular shape.

When the plate is detected, the next process is to crop

the plate containing the character. If the plate is not

detected it will be done detection of the character

directly from the image that is capture with the

webcam.

The third stage is character segmentation, at this

stage using the Connected Component method to

separate the plates and characters. With the connected

component method, each BLOB will be labeled and

filtered by area as well as the size of the blob height

and width. A blob Area smaller than 150 pixels is not

considered a character. Subsequent blob-categorized

characters must be eligible where the vertical height

is greater than the horizontal width. The height and

width ratios should not exceed 0.2 of the height and

width of the captured image.

From the results of the already filtered blob will

be selected eight BLOB ones whose positions are

most parallel. Then the character is sorted into an

array shape and will be compared and recognizable

one the character.

The next stage uses the K-Nearest Neighbor

(KNN) method for character recognition. Each

character that has been separated will be matched by

this method.

Figure 6: Diagram block system.

Object

Camera

Compute

Object Detection

Results

Two Wheel Vehicle License Plate Detection System with Image Processing

Figure 7: Flowchart system.

3.1 Testing

This test is conducted to find out if the tool can detect

the plates and characters on the license plate of the

vehicle that is tested on the daytime conditions with a

range of Lux value 1000-7000 using the Lux meter.

The data retrieval process on this test is done as

follows:

Vehicles made for testing are motors.

Maximum image capture distance of 1meter.

Sampling Place: Outdoor (Campus Park)

Taken with a webcam.

4 RESULTS AND DISCUSSION

This research is conducted with several testing stages

i.e. sample image capture testing based on daytime

conditions and distance of retrievals, vehicle plate

detection testing, character segmentation testing, and

plate character recognition testing.

Figure 8: Sampling object.

4.1 Image Sampling Testing

At sampling testing was conducted by taking an

image with daylight conditions using a webcam with

a resolution of 1600 x 896 pixels and the test distance

performed on this sampling is a maximum of 1 meter.

4.2 Vehicle Plate Detection Test Result

On the plate, detection testing conducted trials with 5

different vehicles as presented in Table 1. The results

of the number plate detection test that successfully

detected the platform precisely from 5 samples in

Table I are as many as 4 samples. In the detected plate

will appear as a green rectangle on the sample image.

License plates can be detected if the plate has an

unbroken outline and has a rectangular contour that is

clear as Figure 11. As for the sample that failed to be

caused by the discovery of the contour that connects

and rectangular shape in the image, as shown in

Figure 12.

Figure 9: Plate sample 1.

Figure 10: Plate sample 2.

ICAE 2020 - The International Conference on Applied Engineering

Table 1: Character recognition results.

Figure 11: Detectable plates.

Figure 12: Undetectable plates.

4.3 Character Segmentation Testing

Results

Segmentation is done with 2 cases. First, if the plate

is detected, the character will be detected using crop

plate results as input in the character detection

function. For the first case, the results are obtained as

can be seen in Figure 13-15/

second case is if not detected plate or plate

detected has no character inside of it as shown in

Figure 14. In this case, the character segmentation

will appear using the original image.

(a)

(b)

(c)

Figure 13: (a), (b), (c) Detectable plates.

Figure 14: Undetectable plates.

Figure 15: Character segmentation.

4.4 Character Recognition Testing

results

In the testing of the character recognition test with

some samples such as Table 2. The test result of Table

2 obtained an accuracy rate of 87.5% from 64

characters. For characters that fail to be detected due

to several factors such as joining the Blob of the

character with another object blob, the detected letters

are too bold compared to the trained dataset.

Two Wheel Vehicle License Plate Detection System with Image Processing

Table 2: Character Recognition Results

5 CONCLUSIONS

Based on the test results of the motor plate detection

of the success rate depends on the shape of the

rectangular contour found, when the plate image has

a clear contour line and not cut off, then the number

plate will be detected accurately. Also, the light

affects the image quality results captured by the

camera. The higher the light then the more noise is

detected.

The results of the character segmentation trial

success level are already quite accurate using the

connected component method. With an accurate rate

of about 80- 90% in separating part of the character

part on the number plate.

For the results of the character recognition trials

with the method K-Nearest neighbour level of

accuracy in 80 characters is 84.5%. Character

recognition using the K- Nearest neighbour method in

a proven application can be applied. Although overall

the success rate of 10 samples of motors taken with

changing light conditions with the assumption of

errors that can be tolerated as much as one character.

REFERENCES

Babu, K. M., Raghunadh, M. V., 2016. Vehicle number

plate detection and recognition using bounding box

method. In International Conference on Advanced

Communication Control and Computing Technologies

(ICACCCT), Ramanathapuram, pp. 106-110. IEEE.

Budianto, A., Adji T. B., Hartanto, R., 2015. Deteksi

Nomor Kendaraan Dengan Metode Connected

Component Dan SVM. Journal TIM Darmajaya.

Kusumanto, A. N. T. R. D., 2011. Pengolahan Citra Digital

Untuk Mendeteksi Obyek Menggunakan Pengolahan

Warna Model Normalisasi Rgb. In Seminar Nasional

Teknologi Informasi & Komunikasi Terapan..

Nur Taufiq, A. H. R. I. M 2012. Sistem Pengenalan Plat

Nomor Polisi Kendaraan Bermotor Dengan

Menggunakan Metode Jaringan Saraf Tiruan

Perambatan Balik, Universitas Diponegoro: Electro

UNDIP.

Qadri, M. T., Asif, M., 2009. Automatic Number Plate

Recognition System for Vehicle Identification Using

Optical Character Recognition. In International

Conference on Education Technology and Computer,

Singapore, 2009, pp. 335-338, IEEE.

Ruslianto, A. H. I., 2011. Pengenalan Karakter Plat Nomor

Mobil Secara. Indonesian Journal of Electronics and

Instrumentation Sytems.

Santra, S., Roy, S., Sardar, P., Deyasi, A., 2019. Real-Time

Vehicle Detection from Captured Images. In

International Conference on Opto-Electronics and

Applied Optics (Optronix), Kolkata, India, 2019, pp. 1-

4. IEEE.

Sari, S. N. D., Fadlil, A., 2014. Sistem Identifikasi Citra

Jahe Menggunakan Jarak CzekanowskI. Jurnal Sarjana

Teknik Informatika..

Setiawan, C. P. H. M., 2015. Pengenalan Plat Nomor

Kendaraan Menggunakan Neural Network. Tugas

Akhir, Politeknik Negeri Batam.

Tauchid, N. A., Rumani, R., Irawan, B., 2015. Analisis

Performansi Metode KNN (K-Nearest Neighbor)

Untuk Pengenalan Karakter Pada Plat Nomor

Kendaraan Di Raspberry Pi. Jurnal e- Proceeding of

Engineering.

Triyandil, J. A. D., 2014. Sistem Otomatisasi Gerbang

Dengan Pengolahan Citra Membaca Nomor Plat

Kendaraan, Skripsi Universitas Komputer Indonesia.,

Bandung, jawa barat, Universitas Komputer Indonesia.

ICAE 2020 - The International Conference on Applied Engineering