Anda di halaman 1dari 17

A Real-Time Recognition System for

Saudi License Plates using LabVIEW


By:
Abdulrahman Al-Juraifani
Yazeed Al-Audah
Ahmad Al-Zuhair
Level:
Undergraduate

Advisor:
Dr. Mohamed Deriche

13/2/2012

Abstract Automatic License Plate Recognition (ALPR) technology covers


numerous applications in traffic monitoring and access control. Several LPR
methods have been presented over the last few years, many of which exhibit
large computational load making them slow and impractical for real-time
applications. In this work, we propose a real-time license plate recognition
system that we simulated using LabVIEW. The system discussed here has
been designed to work with Saudi Arabian license plates and achieves realtime performance. To reduce complexity, the proposed system uses a hybrid
approach for locating the license plate region and recognizing the characters.
The system utilizes low resolution images of 640x480 pixels in order to
further enhance the processing speed and reduce the size of the stored data.
The success rate can reach, under optimum conditions1, up to 94%, with an
average success rate of 84%; with a processing speed of less than 40
ms/plate which is less than half the processing time achieved by other
authors. The system is fully functional and ready for deployment in
commercial systems.

Optimum conditions: good lighting, nice weather and clean plate.

A Real-Time Recognition System for Saudi


License Plates using LabVIEW
TableofContents
1.

Introduction .......................................................................................................................... 4

2.

Background .......................................................................................................................... 5

3.

Proposed System ................................................................................................................. 6


3.1 License Plate Localization .............................................................................................. 7
3.2

LP Segmentation & OCR ........................................................................................ 11

3.2.1 Preprocessing........................................................................................................... 11
3.2.2 Character Segmentation & Recognition ............................................................ 12
4.

Results ................................................................................................................................. 15

5.

Conclusion.......................................................................................................................... 16

References ................................................................................................................................... 17

ListofFigures
Figure 1: A typical Saudi Arabian license plate .................................................................... 5
Figure 2: Localization phase block diagrams......................................................................... 7
Figure 3: Original acquired image and its masked greyscale image ................................. 8
Figure 4: Result of : a) Background Correction Local Threshold ..................................... 8
Figure 5: Result of Particle Filter and Centroid .................................................................... 9
Figure 6: LabVIEW block diagrams generated form Vision Assistant that gives License
Plate Center .................................................................................................................................. 10
Figure 7: processing the image before the OCR phase ...................................................... 11
Figure 8: Results of: a) Niblack Local Threshold,
b) Remove Border Objects
c) Dilation,
d) Gaussian filter & Remove Small Objects ..................... 12
Figure 9: The OCR Training Session ................................................................................... 12
Figure 10: Complete Representation of Center, Masking operation and the OCR block
diagrams of LabView ................................................................................................................ 13
Figure 11 : User interface showing final results during the day and night .................... 14

1.

Introduction

The field of Image processing has allowed many applications to arise


since its beginnings in the 1960s. One of these applications is the
recognition of license plates. License plate recognition (LPR) plays an
important role in many applications such as traffic monitoring and
controlled parking, as well as other applications such as automated toll
collection and security of restricted areas, and so on. Many of these
applications require real-time performance to be practical.
Unfortunately,

most

existing

systems

exhibit

complex

high

computational load, making them inefficient and slow. Other systems use
cloud computing in order to process the images. Using slow and
complicated systems can increase the operational costs of such systems
greatly.
This brings up the need for efficient and powerful digital image
processing techniques and algorithms to be developed.
Numerous approaches to LPR have been proposed in the literature.
However, only a few papers have studied the performance of a real-time
LPR system, and even fewer papers study the recognition of bilingual
license plates such as most GCC license plates. In addition, existing
commercial systems that are able to recognize Saudi or GCC license
plates are proprietary. Thus, small businesses and individuals are unable
to access this technology.
It is important that an open-source, fast, and efficient system be
developed to recognize Saudi license plates.
In this work we approach the problem of real-time LPR by utilizing
the NI LabVIEW software. Our method has proven to be surprisingly fast
compared to other systems, with an average processing speed

of 38 msec/image1, and an average success rate of 84% under different


conditions.

2.

Background

After the initial stage of image acquisition, typical LPR systems


undergo three main phases. Namely, License plate localization,
segmentation, and recognition. This is generally true regardless of the
type of the plate or the alphabet used. License plate localization is a very
essential part in any LPR system.

A typical Saudi Arabian license plate is shown in figure 1. Each license


plate contains three Arabic letters and at most four Indian numerals, and
their corresponding English letters and Arabic numerals.

Figure 1: A typical Saudi Arabian license plate

Many approaches to license plate extraction have been proposed, Hao


Chen et al. [1] have suggested a method based on the texture and edge
information of the plate. Chen et al. [2] and Faradji et al. [3] Proposed
different methods based on morphological techniques. In our system, we
use a basic image processing technique called Local Threshold, as well as
two different morphological techniques to locate the plate. Then we use
1

The processing speed depends on the specifications of the PC running the software. The
specifications under which these results were found are shown in the results section.

LabVIEWs built-in Optical Character Recognition (OCR) tool to


segment and recognize the characters of the license plate.
After the license plate has been located in the image, the next phase is
to segment the characters of the plate, and later, recognize them. For
GCC license plates, the Arabic letters, as in English letters, are in their
separated form, this makes the segmentation phase easier. In our
proposed system, we focus more on segmenting and recognizing the
English letters, since they are more widely used. In this phase, a special
OCR tool in LabVIEW, when set properly, allows us to segment the
characters and recognize them easily and efficiently. This has allowed us
to achieve a very high recognition rate of 99% while still having a very
fast processing speed.

There are many different approaches to character recognition. For


example, Lu et al. [4] proposed a system that uses a word image
matching technique based on an Adaptive Weighted Hausdorff Distance
(AWHD) algorithm to achieve recognition rates up to 93%. In addition,
Liying and Xiaofang [5] used a phase correlation algorithm based on
FFT.

3.

Proposed System

The license plate recognition (LPR) system proposed is based on two


main phases; License plate localization, and segmentation with OCR. As
an initial step, the license plate must be located in order to read the
characters. After locating the plate, masking is performed to extract only
the license plate region out of the whole image. The extracted portion,
which is the license plate, is then forwarded to the segmentation and OCR
phase.

The license plate localization phase is a very crucial stage. Failing to


achieve this task will not allow the reading of the plate to proceed. The
following subsections explain in detail the two main phases of the
system.
3.1 License Plate Localization
The localization phase plays a key role in the overall system. In this
phase, the image will undergo a number of operations as shown in figure
2.2

Figure 2: Localization phase block diagrams

Each block diagram has a specific task and they are cascaded to
produce the license plates center coordinates. First, the block Image
Mask 1, is used to discard the borders of the image that are statistically
unlikely to contain the license plate, it is assumed the license plate is not
on the border of the image. This is done to speed up the overall
processing of the image. Next, Color Plane Extraction 1 is used to
convert the image into a gray-scale one (see figure 3).

To make the design easy, these processes were designed using NI Vision Assistant, and later
transferred to LabVIEW.

Figure 3: Original acquired image and its masked greyscale image

The threshold 1 block is used to convert the gray-scale image into a


binary image in order to easily distinguish between the foreground (plate
characters) and the background (plate background). A backgroundcorrection local threshold is used, since it minimizes the non-uniform
lighting effects, and minimizes the noise in the resulting binary image.
The advanced morphology 1 block removes objects based on their
connectivity to the border (background) of the image (see figure 4).
b)

a)

Figure 4: Result of : a) Background Correction Local Threshold


b) Removing connected objects using Advanced Morphology block

Next, particle filter 1 is used to eliminate any objects that have very
small or very large areas compared to the license plate. This usually
removes all objects in the image except the license plate. Finally, the
8

centroid 1 block diagram returns the center coordinates of the license


plate (see figure 5 & 6). It is worth mentioning that in this preliminary
stage, we do not care much about the clarity of the characters since we
are only after the coordinates of the license plate.

Figure 5: Result of Particle Filter and Centroid

Figure 6: LabVIEW block diagrams that determines the license plates center

10

After obtaining the coordinates of the center, the original image is


automatically masked into a rectangular shaped region around the center.
The mask is designed such that only the English characters and Arabic
numerals are visible.

3.2 LP Segmentation & OCR


OCR is the process by which the system reads characters in an image
after separating them in blocks. The separation process is called
Segmentation. However, before proceeding to the segmentation and
recognition processes, the original image must be prepared or
preprocessed.
3.2.1 Preprocessing

The masked image is passed through the set of processes, shown in figure
7, to prepare it for the segmentation and OCR phase. We call this
preprocessing.

Figure 7: processing the image before the OCR phase

Threshold1 is a Niblack local threshold. Advanced Morphology 1 is used


to remove border objects. Basic Morphology 1 has been used here for
dilation. Dilation is used to fill any gaps in the characters and connect
them. The Dilated image is then fed to a Gaussian filter, Filter 1, for
smoothing and noise reduction prior to the particle filter. Advanced
Morphology 2 is used here to eliminate small objects, relative to the
characters, that are considered as noise. The results of these steps are
shown in figure 8.

11

a)

b)

d)

c)

Figure 8: Results of: a) Niblack Local Threshold,


b) Remove Border Objects c) Dilation,
d) Gaussian filter & Remove Small Objects

3.2.2 Character Segmentation & Recognition


Now that the image is ready, it is forwarded to LabVIEWs OCR tool.
The tool was previously trained on a set of characters and is designed for
the size and spacing of the characters in the plate (see figure 9).

Figure 9: The OCR Training Session

The OCR tool extracts unique features from each segmented object in the
image and compares them to each character stored in the character set. It
then returns the closest character from the character set that best matches
the object and returns a nonzero classification score. The character would
be accepted if its value is higher than the acceptance level [6]. LabVIEW
12

block diagrams that were utilized to perform the previous processes are
shown in figure 10.

Figure 10: Complete Representation of the masking operations, the preprocessing,


and the OCR block diagrams of LabVIEW

13

Results of three samples of the proposed system and their intermediate


phases are shown in figure 11.

Figure 11 : User interface showing final results during the day and night

14

4.

Results

The images used to test the system are all of size 640x480 pixels, and
it is assumed that the license plate is clean, and that the lighting
conditions are good. A sample of 112 images has been used to test the
system. Of these images, 95 where successfully located. This amounts to
a localization success rate 84.8%.
Of the 95 successfully located plates, 94 plates were recognized
correctly. A plate is said to be unsuccessfully recognized if at least one of
its characters was not recognized correctly. Therefore, the success rate for
the recognition with our set of 95 images reached 99%. Combining this
rate with the localization success rate, the overall success rate is around
84%.
It is important to note that we used a set of images to train the
recognition system. Each character was trained 8 times, on average.
Our testing was done on an Intel Xeon CPU running at 2.67 GHz with
4 GB of RAM. Our results show a very fast recognition speed (an
average of 38 ms/plate) which has proven to be much faster than the
results mentioned in table 1.
Table 1: Summary Of Final Results

System

Localization
Success Rate

Recognition
Success Rate

Overall
Success Rate

Average Speed
(s/image)3

Proposed

85%

99%

84%

0.038

[4]

N/A

93%

N/A

0.089

[7]

91.7%

90.9%

83%

1.1

[8]

95.6%

93.7%

87%

N/A

3
The processing speed depends heavily on the system used to run the testing. The numbers mentioned
here are based on different systems. They are mentioned to give an idea about the relative speeds, not
for the sake of comparison.

15

5.

Conclusion

In this paper, we have developed and implemented a system for


license plate recognition using the NIs LabVIEW software. The system
is developed under NI environment so that seamless deployment on other
commercial platforms is possible. The system starts with thresholding
and other morphological techniques to locate the license plate in the
image. A built-in OCR tool for character segmentation and recognition is
then used to identify the LP. A database of more than 100 images was
used to test the system. A success rate of 94% was achieved under
optimum conditions and an overall success rate of 84% with an average
processing time of 38 msec/image. The simplicity of the system and its
low computational load makes it very attractive for real-time
applications. Future work considers improving the localization technique,
and the extension of the system to other GCC license plates.

16

References

[1] H. Chen, J. Ren, H. Tan and J. Wang, "A Novel Method for License Plate
Localization," in Fourth International Conference on Image and Graphics, 2007.
[2] X.-F. Chen, B.-C. Pan and S.-L. Zheng, "A License Plate Localization Mehtod Based
on Region Narrowing," in Proceedings of the Seventh International Conference on Machine
Learning and Cybernetics, Kunming, 2008.
[3] F. Faradji, A. H. Rezaie and M. Ziaratban, "A Morphological Based License Plate
Location," in IEEE International Conference on Image Processing, San Antonio, 2007.
[4] S. Lu, Z. Liu, Y. Chen and L. Liu, "AWHD for License Plate Character," in ICESS,
2008.
[5] L. Liying and Z. Xiaofang, "Application of Phase Correlation Algorithm in Vehicle
License Plate," in ICICTA, 2008.
[6] National Instruments, "NI Vision 2010 Concepts Help," June 2010. [Online].
Available: http://zone.ni.com/reference/en-XX/help/372916J-01/.
[7] P. Comelli, P. Ferragina, M. N. Granieri and F. Stablie, "Optical recognition of motor
vehicle license plates," IEEE Trans. Veh. Technol., vol. 44, no. 2, pp. 790-799, 1995.
[8] H. A. Hegt, R. J. Dela Haye and N. A. Khan, "A high performance license plate
recognition system," IEEE Int. Conf. Syst. Man. Cybern., vol. 5, pp. 4357-4362, 1998.

17

Anda mungkin juga menyukai