Está en la página 1de 5

VEHICLE NUMBER RECOGNITION SYSTEM FOR AUTOMATIC

TOLL TAX COLLECTION


Shoaib Rehman Soomro Mohammad Arslan Javed Fahad Ahmed Memon
TE-08, Electrical (Telecommunication) Engineering Department, Sukkur IBA

Abstract Vehicle Number Recognition (VNR) is an image proposed system consists of five steps. (1) Detection of the
processing technology which uses efficient algorithms to detect vehicle & capturing the image of front view of vehicle. (2)
the vehicle number from real time images. The objective is to Extract & localizing number plate area using vertical edging.
design an efficient Vehicle Number Recognition System & to (3) Number plate segmentation & character separation. (4)
implement it for automatic toll tax collection. The system detects Template matching using correlation to convert characters of
the vehicle first and then captures the image of the front view of
pixel value to alphanumeric value. (5) Using detected license
the vehicle. Vehicle number plate is localized & characters are
segmented. The system is designed for grayscale images so it number to charge toll tax accordingly, storing details in
detects the number plate regardless of its color. Template database, receipt generating & communicating with hardware
matching technique is used for character recognition. The to automate road barriers.
resulting vehicle number is then compared with the available
database of all the vehicles so as to come up with information This research work is ordered as follow; section 2 reviews the
about the vehicle type & to charge toll tax accordingly. The literature of previous research, section 3 presents the proposed
system is then allowed to open road barrier for the vehicle & VNR system, Section 4 briefly discusses the hardware &
generate toll tax receipt. The vehicle information (such as passing database part to make an efficient automated toll tax collection
time, date, toll amount) is also stored in the database to maintain
system, Sections 5 and 6 describe experimental results,
the record. The hardware & software integrated system is
implemented & a working prototype model is developed. conclusion and future work respectively.
Experiments show that system successfully detects and recognize
the vehicle number plate of real images of Pakistani vehicles.
II. LITERATURE REVIEW
Keywords: Vehicle Number Recognition; Automatic Toll Tax Typical VNR System consists of four modules: image
Collection; AVNR; VNI; ANPR
acquisition, license plate extraction, character segmentation,
.
and character recognition. The efficiency & accuracy of the
system largely depends on the second module & various
I. INTRODUCTION approaches have been used for this purpose.
Vehicle Number Recognition (VNR) also known as Automatic There are several common searching algorithms to locate
Number Plate Recognition (ANPR) was invented in 1976. vehicle license plate. Searching algorithm rely on color
Many scientist groups took interest in VNR after 1990s with information [2]. In this method a color search algorithm is
the development of digital camera and the increase in used to extract the likelihood ROI in an image [2]. These
processing speed. VNR is an image processing technology algorithms are usually fast but can detect only single colored
which enables to extract vehicle license number form digital standardized number plate. High license plate extraction rate is
images. It consists of a still or video camera which takes the achieved in [5], [6] based on vertical edging and mathematical
image of vehicle, find the location of the number in the image morphology operations; because of having vertical edges in
and then segments the characters & by using the template English characters & digits, they can be easily classified.
matching scheme, it translates the license number of pixel Several algorithms also utilize neural networks for license
value into numerical or string. VNR can be used in many areas plate extraction [4].
from speed enforcement and motorways to automation of There are also some algorithms designed for recognizing the
parking lots, etc [1].It can also be used on highways & number plates of Pakistani vehicles [2][5]. The system [2]
motorways to automate the toll tax collections. The system utilizes color searching algorithms and effectively detects the
proposed through this work is efficient for automatic toll tax number plates of Sindh only. The other systems [5] rely on the
collection using Vehicle Number Recognition System. The width to height ratio of the standard number plate & matches
earlier methods use plate color information which can detect input vertical edges with that ratio for extracting number
only single color number plates or use specific color search plates.
algorithm which is computationally expensive or use artificial Presently, there are several common algorithms for the
neural network which involves complex mathematics segmentation of license plate characters, such as segmentation
[2][3][4]. The proposed VNR system is efficient & color through dilation, template matching & projection analysis. In
independent so that it can run real time using normal desktop the segmentation through dilation, characters of number plate
PC and can recognize various standard number plates such as are dilated vertically for separating each character & smearing
Sindh (Yellow), Punjab (Green & white), Government (Green) algorithm is used for finding character region [7], the license
& Islamabad (white) under acceptable lighting conditions. The plate characters are also segmented by drawing vertical

978-1-4673-4886-7/12/$31.00 2012 IEEE 125


projection of number plate & finding thee region of each work only vertical mask of Sobel iss used. If the image is not
character [8]. This algorithm is simple andd rapid, but can very much skewed, vertical edge detection provides good
create problem if the plate have dots or imagee is noisy. results [5]. The result of edge detection is depicted in figure 4.

III. PROPOSED VNR SYSTEM


M
Proposed VNR system has four main moduules as shown in
Figure 1. In the first module, it gets inpuut image through
average digital camera. In second module,, license plate is
localized using vertical edging technique. In the third module,
characters of extracted plate are separatedd using character
segmentation. Finally characters & numberrs are indentified
using template matching. Each module contains several
Figure 3. Grayscale Image Figurre 4. Image with vertical edges
processing steps, a flow chart for the propoosed algorithm is
shown in Figure 2.
C. Histogram Analysis of Bit Den nsity
In this step, bit density of rows in figure-4
f is examined using
Image histogram analysis. For this purp pose, number of ones is
Acquisition calculated for each row & presented through vertical
projection. Vertical projection is a graph with two axes.
Vertical axis shows the rows of the image, and horizontal axis
depicts the number of white pixels in each row. Figure 5
shows the vertical projection of the edged
e image.
Number Plate
Extraction

Character
Segmentation

Figure 5.Vertical Projection of bit density

Figure-5 shows that the number platte has the highest values in
Template the vertical projection. Therefore, the
t next step is to find the
Matching rows with the a% highest values in the projection. These rows
are the candidate regions for the plate. Next steps will be
m this step. Figure 6 shows
applied on the results obtained from
Figure 1. Modules of VNR System Figure 2. Proposed Algorithm Flow the resulted image.

A. Image Acquisition
First of all, an image is captured from the front of vehicle
m, for simulation
using a digital camera. In the present system
results, images are taken using a Nokia phonne (5130) having
resolution 300x400 pixels & for the prototypee model, webcam
is used. The captured image is then acquired in Matlab & then
converted into grayscale. Figure 3 shows thee image captured
& converted in grayscale.
B. Vertical Edging
This step explores the property of Englissh Characters &
Figure 6. Result of Bit Density
D Analysis
Digits, because of the digits & characters, image will have
sharp edges in the number plate area [6]. TThis property has
been used for extracting & locating num mber plate within
D. Row Deviation
image. Several methods have been proposed for edge
detection. But research work shows that Sobeel mask has good Headlights & background of the vehicle
v (such as trees) can
performance compared with others [6]. The SSobel method has have vertical edges as well; thereforre they can also be selected
two masks: horizontal mask and vertical m mask. In present as candidate regions in the previou us step. Row deviation is

978-1-4673-4886-7/12/$31.00 2012 IEEE 126


found to prevent this. In Pakistan, number pplates are usually
located in middle or offset on the vehicless bumper. Also a
Number plate has vertical edges in a narrow w range, but those
background & headlights have vertical edges in an wide range.
This feature is used in row deviation [6]. B By calculating the
deviation quantity in each row of the verticaal edges, it can be
easily classified. An easy method for doing tthis is to find the
number of deviations between ones and zeros for each
row within it 80% middle columns of figure--6. If the row has
b% minimum deviation means this row bellongs to plate, so
other rows from candidate regions will bbe removed [6]. Figure 9. Dilated Image
Number plate has vertical edges very close, therefore rows in
the plate region will have consequent points which are totally
ones or zeros and have minimum devviation. Figure 7 F. Removing Small Objects & Meedian Filtering
depicts the vertical projection of row deviatio n. Along with Number plate there caan have small lines in the
candidate region. In this step those small
s regions are removed.
All those regions which have pixels less then P-pixels are
removed. In this way there will bee only single object in the
image which is the location of numb ber plate. In the next step, a
1x15 median filter is applied on th he image to make smooth
skeleton of the number plate. The reesult image after removing
small objects & median filtering is shown
s in figure-9.

Figure 7. Row Deviation Analysiis

As already shown & discussed, structure of thhe vehicles is also


considered as candidate region. This is ddue to boundary
structure & background of the vehicle. The fiigure 8 shows the
result obtained after considering deviation anaalysis.

Figure 10. After Filtering

G. Number Plate Extraction


The image in figure 10 is multiplied with original image of
figure 3 & then by applying horizzontal & vertical scanning
number plate is cropped & extracted
d.
H. Plate Segmentation
Figure 8. Result of Row Deviatioon
In this step, the characters & digits of the plate are segmented
E. Dilation and each is saved as different imaage. To do this 1st of all
In this step, morphological operator is useed. The resulting number plate is resized to a speccific size & image region
image of vertical edges is dilated horizontallly in first attempt properties are found and details of o each connected object
and dilated vertically in second attempt. The structuring (character) within extracted image are
a determined. By finding
elements of dilations are 6-pixel horizontal or vertical lines. out the properties, start & ending coordinates of each character
The image is dilated to connect characters of plate & making are found & each character is saved as different image in a cell
skeleton of the number plate. Result of dilaation is shown in of matrices. Figure 12 shows the nu umber plate characters after
figure 9. segmentation.

Figure 11. Extracted Number Plate Fig


gure 12. Segmented Number Plate

978-1-4673-4886-7/12/$31.00 2012 IEEE 127


I. Template Matching developed, the whole system is setup for testing. The proposed
The last step of vehicle number recognition is the template algorithm has been designed in Matlab R2007 for recognition
matching. For matching the characters with stored characters, of vehicles number plates. The system had utilized Compaq
input images must be equal sized with the stored characters. In Evo N610c with 2.00 GHz processor & 512 MB RAM. The
the present work 50x30 pixel characters are used. When the usefulness of the proposed algorithm has been tested over
extracted characters from plate and stored characters are both images captured in various lighting conditions with a 2
equal sized & each input character image is compared with the megapixel camera of mobile phone (Nokia 5130). The image
ones already stored using cross correlation and the best resolution is used as 400x300 pixels for real scenes & for
similarity is measured. In the Pakistani license plates, all 36 prototype model a web camera is used to capture real time
alphanumeric characters (26 alphabets and 10 numerals) are images with resolution of 640x480 pixels. The system takes an
used, therefore this system is used and each input character is average of 3 to 4 seconds from detection of the vehicle to
correlated with all 36 alphanumeric characters. opening road barrier.
In the template matching each input character of the plate is
correlated with all templates, then from the templates that
character is selected which has highest value of correlation VI. CONCLUSIONS & FUTURE WORKS
coefficient with input character. In this paper, we proposed a real time and efficient
method for Vehicle Number recognition & implementation of
that method for automatic toll tax collection. The system has
been tested on many images of various lighting conditions &
IV. INTERFACING WITH HARDWARE MODEL system can be implemented on motorways & highways for
The VNR system is interfaced with hardware model & automatic toll tax collection.
database to make an automated toll tax collection system. The
hardware model consists of proximity sensor to detect the The proposed system works quite well however, there are
presence of vehicle, a web camera to capture the image, still areas for improvement. The camera used in this project is
motors to open/close the road barriers of toll plaza, desktop average quality & cannot detect fast changing targets due to
computer on which VNR algorithm is executed, LCD & seven the long shutter time. The system robustness and speed can be
segments display & a microcontroller for controlling all the increased if high resolution camera is used. The template
components of hardware model. As the vehicle arrives at toll matching method used in this project for the recognition is
plaza, the inductive proximity sensor detects the vehicle and subject to problems when detecting the characters such as 8 &
gives a signal to the PC using parallel port. The camera B or 0 & O. The frequency transformation can be used while
connected to the PC captures the image of front view of the
correlating to improve the number recognition of the system.
vehicle & applies VNR algorithm on the image to recognize
.
the vehicles license number. This number is then used to
charge toll tax & generate receipt containing all the
information of vehicle. Also, all the information such as time, REFERENCES
date, plate number & toll amount is stored in database to [1] The Automatic Number Plate Recognition Tutorial, http://www.anpr-
tutorial.com, Accessed on May-2012.
maintain the record. PC then sends the signal to
microcontroller using parallel port & the road barrier is [2] M. Tahir Qadri, M. Asif Automatic Number Plate Recognition System
opened for a time by driving motors & Please Move Ahead for Vehicle Identification using OCR, International Conference on
Education Technology and Computer, pp 335 338, 2009.
is displayed on the LCD to guide the vehicles. Complete
hardware design of the system is shown in figure 13. [3] V. Swetha, D.R. Sandeep Automatic Authorized Vehicle Recognition
System, Chennai and Dr.MGR University Second International
Conference on Sustainable Energy and Intelligent System (SEISCON),
pp 789 790, 2011.
D Matlab Executing VNR Algorithm
A [4] V. Koval, V. Turchenko, V. Kochan, A. Sachenko, G. Markowsky
T Smart License Plate Recognition System Based on Image Processing
Using Neural Network, IEEE International Workshop on Intelligent
A Microcontroller Data Acquisition and Advanced Computing Systems: Technology and
Camera Receipt Printer
B Applications, pp 123 127, 2003.
A
Proximity Sensor [5] A. Tahir, H. Adnan Habib, M. Fahad Khan License Plate Recognition
S Algorithm for Pakistani License Plates, Canadian Journal on Image
E Road LCD & 7- Processing and Computer Vision Vol. 1, No. 2, pp 30-36, April 2010.
Barriers Segment [6] F. Faradji, A. Hossein Rezaie, M. Ziaratban A Morphological Based
Figure 13. System Hardware Design License Plate Locating System, IEEE International Conference on
Image Processing(ICIP), pp 57-60, 2007.
[7] S. Ozbay, and E. Ercelebi Automatic Vehicle Identification by Plate
V. EXPERIMENTAL RESULTS Recognition World Academy of Science, Engineering and Technology
9, pp 222-225, 2005.
Better performance of the system can be achieved when
percentage values for a, b & P are set according to the quality
of camera & resolution of images. After the system has been

978-1-4673-4886-7/12/$31.00 2012 IEEE 128


[8] Ch. Lakshmi, A.J Rani, K.S Ramakrishna M.Kanti Kiran A Novel
Approach for Indian License Plate Recognition System, International
Journal of Advanced Engineering Science and Technologies (IJAEST)
Vol No. 6, Issue No. 1, pp 10-14, 2010.

978-1-4673-4886-7/12/$31.00 2012 IEEE 129

También podría gustarte