1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun.

5 - 7, 2023,
University of Ilorin, Nigeria

1 DEVELOPMENT OF AN AUTOMATED PLATE NUMBER RECOGNITION AND

2 IDENTIFICATION SYSTEM
3
4 Z. Haruna, U. Abdulmalik*, N. S. Usman, A. Umar, Z. M. Abubakar
5
6 Department of Computer Engineering, Faculty of Engineering, Ahmadu Bello University, Zaria-Nigeria.
7 *
Email: [email protected]
8
9 ABSTRACT
10 In this paper, an automated plate number recognition and identification system was developed to explore
11 problems associated with many successful solutions in the field of computer vision. However, these solutions
12 are typically tuned towards a particular environment due to the variations in the features of plate numbers across
13 the universe. As such, a universal solution would be difficult to realize as the image analysis techniques used in
14 developing this algorithm cannot generate a result with a hundred percent accuracy. Thus, this paper focuses on
15 the development of an automated plate number recognition and identification system optimized for Nigerian
16 plate numbers. The system comprises algorithms written in Python with the OpenCV library, canny edge
17 detection, and feature detection techniques combined with the Tesseract optical character recognition engine to
18 identify the characters on the plate. A Microsoft visual basic was used to develop a user interface for data
19 presentation and MySQL server was used as a database. When a local image was tested using the developed
20 recognition algorithm, the simulation result presented an accuracy of 90% at the localization stage and 70% for
21 character extraction. This shows that the algorithm effectively localizes and extracts the characters on the plate
22 number.
23
24 KEYWORDS: Plate number, OpenCV, canny edge detection, tesseract optical character recognition engine,
25 MySQL.
26
27 1. INTRODUCTION

28 Modern intelligent transportation systems (ITS) rely heavily on vehicles. Plate characters offer a uniform
29 method of identifying any vehicle. Vehicle License Plate Recognition (VLPR) has drawn a lot of interest from
30 scientists. It can be utilized by government organizations to locate or identify stolen cars and to gather
31 information for traffic management and planning. VLPR requires generalization and high accuracy in real-world
32 applications due to its intimate ties to public security. Due to their diverse applications, such as electronic toll
33 collection, highway surveillance, unattended parking lots and security monitoring of restricted areas, urban
34 logistics, and traffic law enforcement, intelligent transport systems (ITS) significantly contribute to the
35 advancement of smart cities (Ugwu et al., 2022). Since each car's license plate is unique, automatic plate
36 detection will have a wide range of uses. For instance, while employing intelligent parking, the entrance gate
37 will only be opened and closed for allowed vehicles. Additionally, in vehicles, speed control operations and
38 management of vehicles through tolls will be implemented (Akbarzadeh et al., 2022). Different models were
39 developed for various hardware configurations based on computational power and low cost using a novel set of
40 neural networks. The suggested model precision, energy efficiency, communication latency, and computing
41 delay are verified using various license plate datasets both during the day and at night as well as in real-time
42 presented (Padmasiri et al., 2022). Implemented automatic number plate recognition technology in police patrol
43 cars to automate access control and security, such as identifying fast vehicles on highways and locating stolen
44 cars in real time (Etomi & Onyishi, 2021). Using an automated license plate recognition (ALPR) system in
45 resource-constrained contexts can help with the construction and management of smart cities as demonstrated
46 by Padmasiri et al. (2022). This stage is heavily reliant on the number plate's specifications/characteristics.
47 These characteristics include symmetry, spatial frequency, shape, color, height, and width. The conditions
48 (lighting, visibility, image skew, and camera quality) under which the image was taken, as well as the
49 characteristics of the image itself, determine the algorithm's capacity to recognize the plate. The majority of
50 ANPR algorithms also include a pre-processing stage that gets the image ready using methods like grey scaling,
51 thresholding, and noise removal. The most popular way to scale color pixels to grey is to simply average their
52 RGB values (Etomi & Onyishi, 2021). It is impossible to overstate the importance of an automatic licensed plate
53 detection system (ALPDS) in the modern world. This is because it is crucial to the development of smart cities'
54 intelligent transportation systems. In both the academic and commercial sectors, there is growing interest in the
55 field of license plate number detection (Ugwu et al., 2022). Thus, this paper aimed at developing automated
56 plate number recognition and identification system that explores different successful solutions in the field of

1
 1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun. 5 - 7, 2023,
University of Ilorin, Nigeria

57 computer vision. The system comprises stages: the input stage (camera for image acquisition), the processing
58 stage (conversion to greyscale, filtering, contour detection, extraction, and recognition), plate number
59 identification, and finally, the ticketing stage from a visual basic database.
60 2. THEORETICAL ANALYSIS

61 2.1 The overview of major keywords in this area of research.

62
63 Li and Wang (2022) presented Python as a powerful dynamic programming language that is free, open-source,
64 and cross-platform. It features a vast number of potent built-in objects, standard libraries, and extension
65 libraries, and it supports a variety of programming paradigms. By directly calling the built-in functions or
66 standard libraries, powerful programming functionalities can be implemented. Students can expand their
67 comprehension of the algorithm and principle by studying the source code, which will help them develop their
68 programming skills and code quality (Li & Wang, 2022). Python is a reasonably simple language to learn.
69 Dome and Sathe (2021) presented OpenCV as a library for many programming tasks with the main goal of
70 offering real-time computer vision. Users can imports image, does image segmentation, and extracts the image
71 (Li & Wang, 2022). Intel's research led to the creation of the OpenCV computer vision library's source code.
72 The majority of image-processing software programs are developed in C/C++, and OpenCV is no exception.
73 This is done to account for computing speed. It is a library built on the IPL that has more than 300 C functions
74 and C++ classes (Image Processing Library). Thakare et al. (2018) presented an optical character recognition
75 (OCR) tool for Python software called Python Tesseract. In other words, it will recognize and read any text that
76 is embedded deeply within an image. Tesseract-OCR is implemented by the Python-Tesseract module in Python
77 to transform the image into text. Utilizing a module to call the specified method, image to a string makes the
78 implementation easier. Dome and Sathe (2021) use optical character recognition (OCR) to identify text
79 contained in a digital image. It is mostly used to transform printed, handwritten, or transcribed text into editable
80 and reusable text data. Alfath et al. (2022) referred to Visual Basic as a programming language, Visual Basic is
81 also a tool for creating Windows-based application programs. A database is a collection of connected data that
82 makes it simpler for activities to access information Wang et al. (2018). Setting the administrator permission
83 enhanced the system's security. It was crucial to develop the database management system since it offered
84 crucial technical support.
85
86 2.2 This section discusses related works done in this area of research.

87 Mujawar and Vadi (2022) proposed a Raspberry Pi-based system for automatically identifying license plates
88 that use image processing. In this system, a Raspberry Pi is coupled to a camera and an LCD circuit. The tool
89 continuously looks for signs of license plates in the incoming camera footage. When it notices a number plate in
90 front of the camera, it analyses the camera input and extracts the number plate component from the image. The
91 extracted image is processed using OCR to get the license plate number. The program then displays the
92 extracted number on an LCD. Mujawar and Vadi (2022) presented a system in which when it notices a number
93 plate in front of the camera, it analyses the camera input and extracts the number plate component from the
94 image. The extracted image is processed using OCR to get the license plate number. Etomi and Onyishi (2021)
95 work on a Raspberry Pi, an automated system that makes use of the OpenCV library and py-tesseract-OCR
96 demonstrated. After the camera takes a picture, the Raspberry Pi converts it to grayscale, applies filters,
97 segments it, and recognizes characters before displaying the number plate on the console and then storing it in a
98 database. Making sure the code is effective and that the computational complexity is reduced while improving
99 flexibility is one of the trickiest parts of creating a portable program. The performance time is also shown, and it
100 shows that there is a notable 3-second delay in the final output throughout the entire procedure. AG (2019)
101 presented an Optical Character Recognition (OCR) on images and OpenCV techniques are used by the current
102 Automatic Number Plate Recognition and Identification system to read and identify vehicle license plates. This
103 work concentrated on license plate localization and extraction from images for use in various applications. It
104 does this by utilizing several Python-developed OpenCV library techniques for image segmentation, from which
105 some image segments were checked for characters. AG (2019) identifies the most likely license plate by using
106 the characters on the plate's associated picture's height, width, aspect ratio, and other characteristics. Using these
107 pixel properties, it was able to remove undesired contour lines and curves that stood out as noise while
108 identifying the actual area of the image that contained the plate number. Following the acquisition of the region,
109 OCR was used to extract the license plate text format using a trained template for various character styles.
110 Akbarzadeh et al. (2022) developed an improved algorithm to carry out the plan. First, we go through many
111 difficult steps. The placement of the plate is determined as the initial stage. In the second phase, an initial
112 improvement to reduce the expected noise using a suitable filter, which is provided by the Gaussian function.
113 The remaining stages of the program are: locating the edge of photographs, enhancing updated photos, and

2
 1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun. 5 - 7, 2023,
University of Ilorin, Nigeria

114 choosing the precise location of the plate. The study's ultimate step eventually entails several stages, such as
115 using a neural network to automatically extract the characters from the plate. Padmasiri et al. (2022) provided an
116 Automated License Plate Recognition (ALPR) solution for edge computing in situations with limited resources,
117 which could enable the creation and administration of smart cities. The accompanying environmental
118 restrictions, such as changing weather, plate variances among areas, vehicle speed, distorted characters, filthy
119 plates, shadow, and reflection, make research on ALPR tough even though it is a well-established area in the
120 field of image processing. Ugwu et al. (2022) deliver a clever technique that uses computer vision to recognize
121 car license numbers. Images of car license plates were used as training data for the system. Following the
122 identification procedure, the trained images were pre-processed using the OpenCV library for image conversion
123 and masking. The training images were initially annotated using the Visual Graphic Generator (VGG)
124 annotation tool. Then, Tesseract-OCR was applied to isolate the texts from the photos. BC et al. (2022)
125 presented a surveillance system that records images of moving targets and reads their license plates, which
126 become a significant area of study this time. The dataset for Indian license plates was used in the study to
127 identify license plates. Kaur et al. (2022) give an example of a successful ALPR system that recognizes
128 characters with a CNN. The system efficiency was improved by applying a combination of pre-processing and
129 morphological techniques to improve input image quality. The system can recognize license plates with several
130 lines, slanted text, and multiple fonts, among other qualities.
131
132 3. MATERIALS AND METHODS

133 In this section, the procedure follow for the development of an automated plate number recognition and
134 identification system was presented, which comprises some of the following interfaces as materials for both the
135 hardware and software such as the programming language using Python, Pycharm IDE, Tesseract-OCR, USB
136 camera, Microsoft visual basic, and SQL server management studio as the database in actualizing the research
137 procedure. The methods used in achieving the objectives of this research are presented in this section as follows:

138 3.1 Implementation of plate number recognition and identification algorithm

139 Plate number recognition and identification algorithm was implemented on the pre-processed images using
140 OpenCV and pytesseract. OpenCV was used for image processing tasks such as image filtering, edge detection,
141 and extraction of contour image while pytesseract was used to extract the characters on the contour image.
142 These tasks were programmed using Python with the aid of Pycharm IDE.

143 a. Local plate number images

144 In this paper, 115 local plate numbers were acquired by capturing images of vehicles with plate numbers using
145 an Android mobile phone with the following specification: 1GBRAM, 2.0Hz*4 processor, 8MP camera,
146 2432x2432 image size. Some of the captured images are shown in Plate I as follows:

147
148 a. Front View b. Rear View
149 Plate I: Sample Local Captured Images
150 The acquired images were pre-processed by reducing the image size to 500x360 and cropping the image to view
151 the front or rear area of the vehicle as shown in Plate II.
152
153
154
155
156
157
158
159

3
 1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun. 5 - 7, 2023,
University of Ilorin, Nigeria

160
161
162
163
164 a. Front View b. Rear View
165 Plate II: Resizing and Cropping of Images
166 The next stage of image pre-processing is image conversion, i.e., converting an image from blue ray to grayscale
167 to reduce the high resolution of the image. This is then followed by the final stage which is the image filtering,
168 i.e., the stage at which noise is removed from the image with the aid of a bilateral filter (Blurring).

169 a. Image localization

170 In this step, the following processes were followed:
171 i. Canny edge detection technique was used to acquire the edges of the vehicle in the image and then
172 used contours to ignore the unwanted contours in the image, such as the windscreen, spotlight, etc.
173 ii. Loop over contours was used to find the rectangle of the plate number.
174 iii. After localizing the plate number, the remaining background is negligible, then we proceed with
175 removing the image except for the plate number contour.

176 b. Extract of character from a plate number

177 The following are the steps implemented for the extraction of characters from the localized plate number
178 i. Character segmentation cropped the image and save it with a new name.
179 ii. The last step is character recognition which reads the plate number character in the image and
180 used a command to print the extracted character in the terminal.
181 The flowchart for the automatic plate number detection and recognition system for the plate number extraction
182 from a captured vehicle image is presented in Figure 3.1 comprising of pre-processing of the captured image;
183 conversion, contours detection, and also the plate number recognition:
184

185
186 Figure 3.1: Vehicle Capture and Plate Number Recognition Algorithm

187 3.2 Development of desktop application and database

188 The design of a desktop software application using Microsoft visual basic and the development of a database
189 using Microsoft SQL Server Management Studio is used in this research, the interface for both the front and
190 back end are shown in the appendix that includes the authentication, vehicle user interface, daily access report,
191 and database report for all entries.
192
193 4. RESULTS AND DISCUSSION

194 4.1 Detection Result

195 The following shows the result obtained when actualizing the plate number detection after a vehicle image is
196 captured during the entrance. The processes below describe the steps for character recognition by capturing the

4
 1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun. 5 - 7, 2023,
University of Ilorin, Nigeria

197 vehicle image and then converting it to greyscale to reduce the image quality, dictating the vehicle edge for
198 contour detection. After the detection of contour, tesseract-ocr was used to detect the character on the vehicle for
199 daily entrance report by the developed program.
200
201
202 i. The original image captured from the vehicle
203 ii. Converting the original image from the actual image
204 grayscale
205 to help
206 reduce the
207 noise
208
209
210
211
212
213
214
215
216
217
218
219
220
221 Plate III: Original Image
222 Plate IV: Grayscale Image
223
224 iii. The image shows edge detection for the vehicle
225 iv. Contour edge detection
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240 Plate V: Canny Edge Image Plate VI: Contour Detection Image
241
242 v. Final result of extracted plate number

243
244 Plate VII: Result from Image
245 From the result in Plate VI, it is evident that the system recorded a good accuracy in plate number localization,
246 while the result in Plate VII shows that the system can recognise 6 characters correctly out of the eight
247 characters on the plate. This translates to 75%-character recognition accuracy

5
 1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun. 5 - 7, 2023,
University of Ilorin, Nigeria

248 4.1.1 Analysis Table

249
250
251
252
253
254 Plate VIII: Result Analysis

255 The above plate shows the local plate number sample tested on the algorithm. The total number of plates used is
256 ten, only seven give correct characters, two of them incorrect characters, and only one does not have character
257 or contour. Also, nine out of ten detect contour which gives us 90% contour detection and 70%-character
258 recognition as the overall system performance.
259 4.2 Result Obtained from the Software Developed for Record Based
260 Below is the interface for the daily entrance report and the database entire report for daily entrance. The
261 program is designed to accept information related to vehicle owners before gaining access to the daily entrance
262 which captures the plate number, time, gate passing by, either staff, student, or visitor, and the security attending
263 you during the entrance.
264 i. Daily entrance verification report
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279 Figure 4.1: Daily Entrance Interface
280
281 ii. Database results obtained from daily report
282
283
284
285
286
287
288
289
290 Figure 4.2: Daily Entrance Report
291 5. CONCLUSION
292 The algorithm developed in this work for the extraction of Nigeria plate numbers proved quite effective and
293 efficient. It had a 90% accuracy for license plate localization but a reading accuracy of 70%. MYSQL server
294 database was used in this study which allows large storage of driver’s particulars and vehicle data. With the
295 output tied to a database, this system can be used as an automated process for managing a large number of
296 vehicle user profiles as well as access control. Further work can be conducted to improve the accuracy of this
297 system, using deep learning (tensor flow). Performance comparison for the captured plate number and the record

6
 1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun. 5 - 7, 2023,
University of Ilorin, Nigeria

298 stored in the database can be automated to make sure that each vehicle must be registered before accessing
299 premises.

300 ACKNOWLEDGEMENT
301 I wish to express my profound gratitude to Almighty Allah; the Creator, Fashioner, without His mercies and
302 favour I would certainly not exist. I would also like to express my profound gratitude to my supervisor Dr.
303 Zaharuddeen Haruna for giving me the opportunity, for his patience, guidance, and careful instruction
304 throughout the research, and a lot of others whom I cannot mention here whose hard work and dedication during
305 this study assisted me with advice and encouragement. I have nothing to offer them but for Allah to reward them
306 abundantly and enable them to succeed in all their endeavours. This acknowledgment would not be complete
307 without expressing my sincere gratitude to my parents, brothers, and sisters for their love, patience,
308 encouragement, and understanding which are the source of my motivation and inspiration throughout my work.
309 May Almighty Allah in his abundance reward you all greatly, and then some more. I must in the final analysis
310 not fail to appreciate all those special individuals, colleagues, family members, and friends whom I cannot
311 mention here that have contributed in one way or another, in the course of my struggle and improved on my
312 qualification. Thank you all.
313 REFERENCES

314 AG, S. F. (2019). Development of portable automatic number plate recognition (ANPR) system on Raspberry
315 Pi. International Journal of Electrical and Computer Engineering, 9(3), 1805.
316 Akbarzadeh, O., Khosravi, M. R., & Alex, L. T. (2022). Design and Matlab simulation of Persian license plate
317 recognition using neural network and image filtering for intelligent transportation systems. ASP
318 Transactions on Pattern Recognition and Intelligent Systems, 2(1), 1-14.
319 Alfath, A. N., Taningngsih, I., & Suharto, E. (2022). Designed an Inactive Archives Management Information
320 System Using Visual Basic 2010 at The Department of Community Empowerment and Village West
321 Java Province. Journal of Applied Engineering and Technological Science (JAETS), 3(2), 105-115.
322 BC, M. H., Pukhrambam, S., Gokakakar, R. R., & Kaveri, B. (2022). YOLOV4 object detection for license
323 plate detection.
324 Dome, S., & Sathe, A. P. (2021). Optical character recognition using Tesseract and classification. Paper
325 presented at the 2021 International Conference on Emerging Smart Computing and Informatics (ESCI).
326 Etomi, E. E., & Onyishi, D. U. (2021). Automated number plate recognition system. Tropical Journal of
327 Science and Technology, 2(1), 38-48.
328 Kaur, P., Kumar, Y., Ahmed, S., Alhumam, A., Singla, R., & Ijaz, M. F. (2022). Automatic license plate
329 recognition system for vehicles using a CNN. CMC-Computers, Materials & Continua, 71(1), 35-50.
330 Li, F., & Wang, L. (2022). Application of Case-based Teaching in Computer Vision with OpenCV and Python.
331 environment, 5, 8.
332 Mujawar, S. S., & Vadi, S. (2022). Raspberry Pi and OpenCV for license plate recognition in real-time.
333 Padmasiri, H., Shashirangana, J., Meedeniya, D., Rana, O., & Perera, C. (2022). Automated license plate
334 recognition for resource-constrained environments. Sensors, 22(4), 1434.
335 Thakare, S., Kamble, A., Thengne, V., & Kamble, U. (2018). Document Segmentation and Language
336 Translation Using Tesseract-OCR. Paper presented at the 2018 IEEE 13th International Conference on
337 Industrial and Information Systems (ICIIS).
338 Ugwu, E. M., Taylor, O. E., & Nwiabu, N. D. (2022). An Improved Visual Attention Model for Automated
339 Vehicle License Plate Number Recognition Using Computer Vision. European Journal of Artificial
340 Intelligence and Machine Learning, 1(3), 15-21.
341 Wang, Y., Liu, J., He, X., & Wang, B. (2018). Design and realization of rock salt gas storage database
342 management system based on SQL Server. Petroleum, 4(4), 466-472.
343
344 APPENDICES

345
346
347
348
349

7
 1st Faculty of Engineering and Technology Conference (FETiCON 2023), Jun. 5 - 7, 2023,
University of Ilorin, Nigeria

350
351
352

353

354

355

356

357

358

359
360
361
362
363
364
365
366