Sunteți pe pagina 1din 5

Vehicle license plate detection and recognition using

faster edge detection algorithm


Sakhare Varsha D.

Prof.Mrs.R.S.Nipanikar

Student E & TC Department


PVPIT Bavdhan,Pune. Savitribai Phule University
Pune, India
s.varsha15@gmail.com

Assistant Professor E& TC department.


PVPIT Bavdhan,Pune. Savitribai Phule University
Pune, India
rohininipanikar@rediffmail.com

Abstract:-This paper proposes a fast method of edge detection


to detect and recognize the vehicle number plate.License plate
detection and recognition plays important role in many
applications and many techniques have been proposed for
it.Here Automatic thresholding using integral image is used
for binarisation, after that unwanted line elimination algorithm
(ULEA) is proposed to enhance the image and then vertical
edge detection algorithm (VEDA) is applied. For character
segmentation Optical code recognition (OCR) technique is
used. Computation time is very less for the proposed method.
Keywords:- adaptive thresholding(AT), License plate
detection(LPD), vertical edge detection algorithms(VEDA),
Median filtering.
I. INTRODUCTION
Vehicle license plate recognition plays important role in
many applications such as automatic toll collection, traffic law
enforcement, vehicle tracking, car parking automation and
many other applications. A typical license plate recognition
(LPR) system consists of three parts.1. License plate detection
2.License plate character segmentation 3.Character recognition.
Among these parts the system accuracy depends on license
plate detection so it is important step in LPR.
There are many methods of license plate detection (LPD)
however most schemes are restricted by some practical
restriction such as recognition time, image resolution, lighting
conditions, rainy environment, camera cost, complexity etc. In
automatic toll collection cameras are installed to find the
license plate and then matched with database of the computer
to collect the toll automatically. usually for such application
outstanding cameras were used which increases the cost of
hardware. In this paper low resolution web camera is used to
capture the image and remaining process of plate detection will
be done by offline process. Vertical edge extraction and
detection is an important step in vehicle LPDR as it affects the
system accuracy and computation time.
There are many license plate detection methods based on
texture, color, shape, environmental conditions etc. Basic and
easy method of license plate detection is morphology based
LPD [1].Edge extraction method[2] based on texture mainly
exploited the aspect ratio [3] ,uniform distribution of
characters[4]. Due to some lighting conditions, interference
characters and other problems it becomes difficult to detect the

license plate in complex conditions. The LPD algorithm[5]


using color edges has been proposed but it can not be used with
all colors.
In this paper we proposed faster vertical edge based
algorithm with median filtering. we also used some
morphological operations to remove the noise. For character
segmentation we have used optical code recognition technique
(OCR). This technique gives the output with very less
computation time. Our preference is to use low resolution
images which reduce computation time, memory size and cost.
The remaining paper is organized as below. section II
introduces proposed method for vehicle LPD. Section III
introduces vehicle LPR. Section IV shows experimental set up
and conditions and the paper is concluded in section V. This se
I.

PROPOSED METHOD FOR VEHICLE LICENSE


PLATE DETECTION

A. Overview
In this paper color license plate is converted to grayscale
image.Adaptive thresholding is applied on image to get
the binarised image. Integral image is used for adaptive
thresholding.After that unwanted line elimination
algoriyhm (ULEA) is applied to remove the noise.then
vertical edge detection algorithm (VEDA) is used to
extract the vertical edges.Next median filtering and some
morphological operations are used to remove noise.
Finally vehicle license plate is detected in the original
image. The flow chart for license plate detection is shown
in fig. 1.
Color input image of license plate.
Gray image conversion
Adaptive Thresholding using
integral image
Unwanted Line Elimination
Algorithm (ULEA)

Vertical Edge Detection Algorithm


(VEDA)

The next process is to obtain intensity summation for each


window using two subtractions and one addition. For this
purpose we need only 4 values of integral image matrix.we
will use top right 4 valued rectangle for calculation.

Median filtering
Morphological Operations
Display license plate
Fig.1 Flowchart for LPD

B. Adaptive thresholding using integral images


Adaptive Thresholding is the form of thresholding in
which threshold window moves successively over every pixel
in image. There are many adaptive thresholding techniques
such as AT by using mean, median, integral image,Gaussian
filter. In this paper we are using adaptive thresholding using
integral image[6]. Bradely and Roth [7] proposed real time AT
using mean local window and this window is computed on
integral image.
The AT technique used here is just a simple extension of
Bradley and Roths[7] and Wellners methods [8]. The idea in
Wellners algorithm is that the comparison of pixel is done
with an average of neighboring pixels. An approximate
moving average of the last S pixels is calculated. If the value
of the current pixel is T percent lower than the average, then it
is set to black; otherwise, it is set to white. This technique is
useful because comparing a pixel to the average of
neighboring pixels will keep hard contrast lines and ignore
soft gradient changes.However,Wellners algorithm depends
on the order of scanning.The nighbourhood samples are not
evenly distributed so this process is somewhat unsuitable to
get good representation for neighbouring pixels. The AT
process using integral image [7] can solve this problem.
An integral image is a tool that can be used to when we
have functions from pixel to real numbers. To compute the
integral image we add left and top pixel values of the current
matrix and that addition is replaced by current pixel value.
Image matrix below shows the computation of integral image.

Sum of all values upto A,B,C,D are 6,10,8,16.


SUMwindow = 8
Threshold for each pixel can be calculated as ,
t (i,j) = (1-T)* SUMwindow
Wellner uses one eighth of image width for value of
selection S thus output threshold can be given as,
Output threshold value
O(i,j) =

0; current pixel values < t(i , j)


255 otherwise

Fig. 2(a) shows input image and fig. 2(b) shows output image
after AT.
C. ULEA
ULEA is used to remove many thin lines produced due to
thresholding. Unwanted lines can be formed in four cases;
horizontal with 0 angle, vertical with 90, line inclined
with 45,line inclined with 135. A 3*3 mask is used for testing
and this testing is done on only black pixels. Once, the current
pixel value at the mask is black, the eight neighbor pixel values
are tested. The current pixel value is converted to white only if
two surrounding values of pixel are white. Fig. 3 represents
how the mask moves through image. Fig. 4 shows output
image after ULEA is performed.

[ ]
12 3 4
0 11 4
3104
2 13 1

Simple input image matrix


The computed integral image matrix is,

[ ]
1 3 610
1 4 8 16
4 8 9 23
611 18 30

Fig. 2 (a) Input image

Fig. 2(b) Threshold image

II.

CLP RECOGNITION

In this paper Optical Code Recognition (OCR) technique


is used for character segmentation.The OCR is recognition
method in which the input is an image and we get string of
character as output.The cropped image is compared with the
template data stored in database. Template matching is one of
the important step in character recognition as it affects the
Fig. 3 Cases of ULEA

Fig.7 Detected license plate

Fig. 4 ULEA output image

accuracy. This method overcomes the problem of different


light strength in different time.
III.

EXPERIMENTAL SETUP AND ENVIRONMENTAL


CONDITIONS

The implementation steps are as below.

Fig. 5 VEDA output

1.

Web camera is set to active profile and connected to


laptop.

2.

Web camera is focused on car license plate.

3.

Distance between camera and car license plate is 1-2


meter.

4.

Captured samples contain different background and


objects.
While performing the experiment weather is sunny and
shady days.
IV.

Fig. 6 Design of proposed mask for VEDA

D. VEDA
After eliminating the unwanted lines in the binary image
the Vertical Edge Detection Algorithm (VEDA) is used to
distinguish the plate detail region. The idea of the VEDA
concentrates on intersections of black white region. Basically,
the proposed mask consists of 3 smaller masks, which are left mask
2X1, center mask 2X2, and right mask 2X1

E. MEDIAN FILTERING AND MORPHOLOGICAL


OPERATIONS:
Output of VEDA is filtered by using median filter to
remove spikes in the image. Some morphological
operations were done to enhance the image.Vehicle license
plate is detected after this process. Fig. 7 shows the output
image.

CONCLUSION

We have proposed fast algorithm for vehicle number plate


detection and recognition. The computation time is very less
of this proposed method. In this paper adaptive thresholding
using integral image is used to overcome the problem of
illumination, noise, resolution etc.
a.

REFERENCES
[1] J.W.Hsieh [10] J.-W. Hsieh, S.-H. Yu, and Y. S. Chen,
Morphology-based license plate detection from complex in
Proc. 16th Int. Conf. Pattern Recognit.,Quebec City, QC, Canada,
2002, pp. 176179.
[2] J.-M. Guo and Y.-F. Liu, License plate localization and character
segmentation with feedback self-learning and hybrid binarization
techniques,IEEE Trans. Veh. Technol., vol. 57, no. 3, pp. 14171424,
May 2008.
[3] M. Yu and Y. D. Kim, An approach to Korean license plate recognition
based on vertical edge matching, in Proc. IEEE Int. Conf. Syst., Man,
Cybern., 2000, vol. 4, pp. 29752980.
[4] C. T. Hsieh, Y. S. Juan, and K. M. Hung, Multiple license
plate

detection for complex background, in Proc. IEEE 19th


Int. Conf. AINA,
2005, pp. 389392.
[5] S.-L. Chang, L.-S. Chen, Y.-C. Chung, and S.-W. Chen,
Automatic
license plate recognition, IEEE Trans. Intell. Transp.
Syst., vol. 5, no. 1,
pp. 4253, Mar. 2004.
[6]

Abbas M. Al-Ghaili, SyamsiahMashohor, Abdul RahmanRamli, and


Alyani Ismai Vertical-Edge-Based Car-License-Plate Detection
Method IEEETrans. Veh. Technol., vol. 62, no. 1, January 2013

[7] D. Bradley and G. Roth, Adaptive thresholding using the integral

[1]
image,J. Graph. Tools, vol. 12, no. 2, pp. 1321, Jun. 2007.
[2]
[3] [8] Z.-X. Chen, Y.-L. Cheng, F.-L. Chang, and G.-Y. Wang, Automatic
[4]
license-plate location and recognition based on feature salience, IEEE
[5]
Trans. Veh. Technol., vol. 58, no. 7, pp. 37813785, Sep. 2009.
[6]
[7] [9] T. Naito, T. Tsukada, K. Yamada, K. Kozuka, and S. Yamamoto, Robust
[8]
license-plate recognition method for passing vehicles under outside
[9]
environment, IEEE Trans. Veh. Technol., vol. 49, no. 6, pp. 23092319,
[10]
Nov. 2000.
[11]

S-ar putea să vă placă și