Українська   Русский
DonNTU   Masters' portal

Abstract

Content

Introduction

In our time, characterized by the mass introduction of information technologies in various areas of human life. The development and use of intelligent transport systems is carried out to increase traffic safety and improve traffic conditions.

The problem of recognition of the vehicle license plate is very interesting and for many years attracts many researchers and experts of machine vision. Application of such a system is very high and can vary from the parking lot to the PPC traffic management.

Solution of the license plate recognition can be represented as the complex algorithms of image processing and analysis. All existing methods of identifying characters always start with the procedure of the pre-treatment image. Proper pre-treatment steps, such as binarization, morphological standardization and segmentation, are crucial for the next steps of recognition.

1. Theme urgency

The urgency of the problem of determining vehicle license plates is that nowadays widely developed range of automated traffic control vehicles, monitoring of entry to the company, the rate of movement, as well as to identify violators of traffic rules. Additionally, separate algorithms and methods used for the recognition of numbers, can get a synthesis and widely used in other problems associated with digital processing and analysis of digital images.

2. Goal and tasks of the research

The main aim of the work is the modification and development of algorithms that are part of the complex problem of determining the number plate of the vehicle for a more efficient result.

To achieve this goal, the following tasks:

  1. The analysis methods for determining the license plate area on the input image.
  2. The normalization and segmentation of the license plate.
  3. The method implementation of character recognition and evaluation of its effectiveness.

During the execution of the master's work is planned to examine the existing methods and algorithms for recognition of license plates of transport, as well as to develop and implement an automated system which solves the above problem.

3. Review of research and development

Popularity determining automobile rates can be explained by the fact that the interest in such a software system increasingly began to show a variety of commercial structures for the organization of parking places and control visits to the territory of large enterprises.

It offers a great number of different solutions to the problem of identification of license plates and their recognition. For example, Tran Duc Duan, Duong Anh Duc and Tran Le Hong Du [1] and Remus Brad [2] in his works used Hough transform, in order to reduce the processing time for segmentation of the license plate in the image. Also, in order to achieve faster processing in some systems used thresholding size license plate characters and regions in it.

N. Zimic, J. Ficzko, M. Mraz, J. Virant [3], Eun Ryung Lee, Pyeoung Kee Kim, and Hang Joon Kim [4] and Opas Chutatape, Li Li, and Qian Xiaodong [5] In his works using segmentation algorithms regions characters and their recognition by fuzzy logic and neural networks.

In the paper Wenjing Jia, Huaifeng Zhang and Xiangjian He [6] used an algorithm that determines multiple candidate regions in the image using functions such as rectangular, proportions and the density boundary to determine whether the region identified by the license plate or not.

All methods used in the above-mentioned research activities, seek to maintain the right balance between accuracy and speed of the algorithm.

The basis of any classification are certain classification features (principles). So in the manual [ 7 ] as a qualifying characteristic properties are used information that used in the recognition process. However, this classification does not reflect all the characteristics of real-time systems (RT), and therefore has developed a classification of properties of control systems for pattern recognition (SPR) in the recognition process (Fig. 1).

Классификация СРО РВ

Figure 1 – Classification SPR RT (animation: 5 frames, 5 cycles of repeating, 165 kilobytes)

Consider the most popular system of automatic license plate recognition:

  1. ATAPY ANPR SDK – specialized software with technology of automatic recognition of car number [8]. It designed for system integrators and application developers who want to incorporate technology of automatic recognition of vehicle number in their applications. System ATAPY ANPR SDK has a high recognition accuracy t. To. Worked on one of the finest products in the industry OCR – engine OCR, designed ABBYY Software House [9]. It has a function of automatic skew correction of geometric/tilt (up to 30°).
  2. АВТОМАРШАЛ – license plate recognition system, which has 12 configurations (car wash, parking, storage, PPC and others.) that allow you to easily customize the system to the task [10]. It has user-friendly software for quick setup assistant also supports integration with a variety of equipment (barriers, traffic lights, scales, etc.).
  3. CarGo Enterprise – recognition of all types of Ukrainian and foreign numbers with a probability of 96%. The software recognition system allows numbers to effectively solve the problem of registration, identification, prevention of unauthorized travel, security vehicles, traffic control [11].

The analysis of existing systems, ANPR, the following was revealed:

  1. Complete license plate recognition system not only have a ready solution, but also have great value;
  2. Almost all systems have a large number of optional features such as automatic control barriers and traffic lights, or the formation of reports on various criteria;
  3. All the algorithms and methods used in solving the problem, not available for public viewing;
  4. Most systems support the recognition of not only the local license plates, but also abroad;
  5. Complete systems have a high recognition accuracy.

In the master's work O. V. Zhivotchenko [12] the basic steps and techniques license plate recognition. In the I. S. Lichkanenko [13] consider problems arising in the process of recognition of license plates of vehicles are defined methods of image processing and pattern recognition to identify license plates. In the A. V. Fedorov [14] problem recognition is carried out only by means of neural networks. In the above studies, in contrast to my use other detection methods are not offered new modified methods.

Based on the analysis of existing systems, methods and algorithms used in the development of the system has identified the need to establish a new system of automatic number plate recognition vehicle, with the aim of finding a balance between speed and accuracy of recognition, as well as minimizing the cost of developing the system.

4. Stages of the vehicle license plate recognition

The gradual process of recognition vehicle license plate is shown in Fig. 2.

The process of the vehicle license plate recognition

Figure 2 – The process of the vehicle license plate recognition (animation: 7 frames, 5 cycles of repeating, 123 kilobytes)

The input data is an image of the vehicle captured by the camera.

Binarization

Original Image license plate transport requires some pre-treatment steps to make the appropriate format for the character recognition. Since for most OCR algorithms require monochrome images, you must first binarizirovat image, converted gray color or in black and white. Fig. 3 shows a comparison between the original image and the black-and-white one, resulting in the binarization.

The original image and the image after binarization

Figure 3 – The original image and the image after binarization

Normalization

This step is necessary to improve the layout of an output image. Normalization of the license plate image is carried out in two stages. At the first stage the angle of rotation of the hotel in the image plane. The second – runs an algorithm for obtaining a normalized image number, taking into account the angle of its rotation. To rotate the image area using an algorithm based on the corresponding affine coordinate transformation. To reduce image distortion when you turn related to its discrete nature, used a method based on bilinear interpolation of the closest four pixels [15].

At the first stage the operation is performed on the numbers underscore the borders of a linear operator Sobel horizontal boundaries having convolution mask:

The Sobel operator is more sensitive to the direction of the border close to the horizontal, whereby the captured image is well distinguished upper and lower part of the plate, as shown in Fig. 2.

Figure 2 – a) a fragment of the original image with the found position numbers; b) Cut image of the license plate with the expansion of 40% in the vertical direction; c) the results underscore the borders

In the second step calculates density map points found in the parameter space boundaries of spatial coordinates according to Hough transform. The second phase is to determine the equations of the lines corresponding to the upper and lower boundary of the plate. Each point boundary maps creates many lines passing through it, which satisfy the equation:

yi = axi + b,

in the parameter space that corresponds to:

b = – axi + yi.

The weight of lines v(xi, yi) corresponds to the brightness results underscore boundaries (Fig. 2b). Thus, giving weight v direct and guiding them in the space of parameters a and b with a brightness equal to the weight, we obtain an image similar to that shown in Fig. 3[16].

Карта результатов преобразования Хафа

Figure 3 – Map results of Hough transform

Segmentation

For image segmentation algorithm scan lines based on the availability of particular transitions from 1 to 0 and 0 to 1 in the character in the binary image. Thus, the total number of transitions in the character more than the total number of transitions in the other area. There are at least seven characters on the license plate area and each symbol has more than two leaps [17]. You can select ten as the threshold value. If the total number of transitions in a particular line is greater than ten, then this line can be any character. Otherwise, it is not in character.

Algorithm:

  1. let Н – the height and W – the width of the image plate.
  2. for (i=H/2 to 0) (do not consider the transitions, from 0 to 1 and from 1 to 0 in cnt; if cnt<10 get the coordinates у in ymin and interrupt;)
  3. for (i=H/2+1 to H-1) (do not consider the transitions, from to до 1 and from 1 to 0 in cnt; if CNT<10 get the coordinates у in ymax and interrupt;)
  4. cropped image from ymin to ymax.

The number after the scan line algorithm

Figure 4 – The number after the scan line algorithm

Will review the cropped image from left to right in the columns after the exact location of the upper and lower limits, and calculate the total number of black dots in each column. The threshold value is set to h / 10. We define each value in the array of projections. If the projection of [i] is greater than h / 10, the projection [i] is equal to one. Otherwise, the projection [i] is set to zero. Where h does not modify the line in the binary image after pinpointing the upper and lower bounds [17]. The symbols are then trimmed by selecting parts with projection[i] = 1.

Before recognition algorithm characters must be specified in a block that does not contain extra spaces (pixels) on all four sides of the character.

Characters segmentation

Figure 5 – Characters segmentation

Characters after removing extra spaces from four sides

Figure 6 – Characters after removing extra spaces from four sides

The character recognition

Templating method includes comparing the similarity between the given set of templates and the input image is normalized, the same size as the template, and then define a certain pattern that produces the greatest similarity.

In this paper a formula used to detect the similarity matching between the models of the two signals by cross-correlation. A function of Matlab "corr2" to calculate the correlation coefficients for each comparison between the test image and the template. In the formula below, Amn – the input image, Bmn It is one of the templates. The function returns the value of compliance r shows how well Amn coincides with Bmn. If one of the correlation coefficient is significantly higher, the input image is identified as the letter or digit.

Following detection of line segmentation and character matching process starts reading the input text from the first line to the bottom line, from left to right, the procedure ensures that each letter and an output layout of each line. Then the final step is to write the words to a text file. For the input image in Fig. 7, the system is able to read the image in the text as shown in the figure below.

Output text

Figure 7 – Output text

Conclusion

As part of the research carried out:

  1. Based on the literature review highlighted problems for the implementation of automatic license plate recognition vehicle. Required to solve this problem is to find the boundaries, filtering, normalization and segmentation, and the most appropriate methods of recognition – correlation, K-nearest and neural networks.
  2. The analysis of existing systems, identified advantages and disadvantages.
  3. Implemented preliminary image processing in order to improve their quality and easy localization of the license plate.

Further studies focused on the following aspects:

  1. A deep analysis of the basic methods of image processing and recognition of license plates of vehicles.
  2. Modification of known image processing techniques to improve the quality determination of the plate.
  3. Development of an automated detection system license plate using image processing techniques.

In writing this essay master's work is not yet complete.

Final completion: December 2015. The full text of work and materials on the topic can be obtained from the author or his manager after that date.

References

  1. Tran Duc Duan, Duong Anh Duc, Tran Le Hong Du. Combining Hough Transform and Contour Algorithm for detecting Vehicles License-Plates. Proceedings of 2004 International Symposium on Intelligent Multimedia, Video and Speech Processing. October 2004 – с. 747-750
  2. Remus Brad. License Plate Recognition System. Computer Science Department Lucian Blaga University, Romania. – 5 с.
  3. N. Zimic, J. Ficzko, M. Mraz, J. Virant. The Fuzzy Logic Approach to the Car Numlber Plate Locating Problem. IEEE. 1997
  4. Eun Ryung Lee, Pyeoung Kee Kim, and Hang Joon Kim. Automatic Recognition of a Car License Plate Using Color Image Processing. IEEE. 1994
  5. Opas Chutatape, Li Li, and Qian Xiaodong. Automatic License Number Extraction and Its Parallel Implementation. IEEE. 1999.
  6. Wenjing Jia, Huaifeng Zhang and Xiangjian He. Mean Shift for Accurate Number Plate Detection. Proceedings of the Third International Conference on Information Technology and Applications (ICITA’05).
  7. А. Л. Горелик, В. А. Скрипкин Методы распознавания. – М.: Высш. шк, 2004. – 261 с.
  8. ATAPY ANPR SDK [Электронный ресурс]. – Режим доступа: http://www.atapy.com/Products/AutomaticNumberPlateRecognitionANPRSDK.aspx
  9. ABBYY Software House [Электронный ресурс]. – Режим доступа: http://www.abbyy.com
  10. АВТОМАРШАЛ [Электронный ресурс]. – Режим доступа: http://avtomarshal.ru
  11. CarGo Enterprise [Электронный ресурс]. – Режим доступа: http://intteks.com.ua/component/content/article?id=552
  12. О. В. Животченко Разработка компьютерной системы обработки изображений с камер видеонаблюдения для определения номерного знака транспортного средства. – Портал магистров ДонНТУ [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2012/fknt/zhivotchenko/diss/index.htm
  13. И. С. Личканенко Исследование методов и поиск эффективного алгоритма для задачи распознавания номерных знаков транспортных средств. – Портал магистров ДонНТУ [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2013/fknt/lichkanenko/diss/index.htm
  14. А. В. Фёдоров Исследование методов контурной сегментации для построения системы оптического распознавания символов. – Портал магистров ДонНТУ [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2010/fknt/fedorov/diss/index.htm
  15. Р. Гонсалес Цифровая обработка изображений / Р. Гонсалес, Р. Вудс – М.: Техносфера, 2006 – 1072 с.
  16. К. В. Мурыгин Нормализация изображения автомобильного номера и сегментация символов для последующего распознавания / К.В. Мурыгин – Штучний інтелект 3, 2010 – 6 с.
  17. Ch.Jaya Lakshmi, Dr.A.Jhansi Rani, Dr.K.Sri Ramakrishna, M.KantiKiran, A Novel Approach for Indian License Plate Recognition System, International Journal Of Advanced Engineering Sciences And Technologies Vol No. 6, Issue No. 1 – с. 10-14.