Automatic Number Plate Recognition

A Technical Seminar Report on Underwater Communication

CHAPTER 1

INTRODUCTIONLaw enforcement agencies throughout the nation are increasingly adopting automated

license plate recognition (ALPR) technologies to enhance their enforcement and investigative

capabilities, expand their collection of relevant data, and expedite the tedious and time

consuming process of manually comparing vehicle license plates with lists of stolen, wanted,

and other vehicles of interest. Police officers, sheriff’s deputies, and other law enforcement

practitioners are often on the lookout for vehicles that have been reported stolen, are wanted

in connection with a crime or traffic violation, are suspected of being involved in criminal or

terrorist activities, are parking violation scofflaws, have failed to maintain current registration

or to comply with statutory insurance requirements, or any of a number of other legitimate

reasons. ALPR systems function to automatically capture an image of the vehicle’s license

plate, transform that image into alphanumeric characters using optical character recognition

or similar software, compare the plate number acquired to one or more databases of vehicles

of interest to law enforcement and other agencies, and to alert the officer when a vehicle of

interest has been observed. The automated capture, analysis, and comparison of vehicle

license plates typically occur within seconds, alerting the officer almost immediately when a

wanted plate is observed. Although the ALPR term includes a specific reference to

“automated,” it should be noted that human intervention is needed insofar as the officer

monitoring the equipment must independently validate that the ALPR system has accurately

“read” the license plate, that the plate observed is issued from the same state as the one in

which it is wanted, and to verify the currency of the alert, i.e., verifying that the reason this

vehicle or the owner was wanted or of interest is still valid.

This National Institute of Justice (NIJ)–supported project was designed to assess and

document ALPR implementation and operational experiences among law enforcement

agencies in the United States, and to identify emerging implementation practices to provide

operational and policy guidance to the field. Several data collection techniques were used to

gather information for this project, including:

1) A survey of law enforcement agencies to assess the scope of the current ALPR

implementation, deployment, and operational uses,

2) Site visits to interview law enforcement practitioners and observe ALPRs system in

operation,

3) Reviewing documents and policies addressing ALPR implementation and use.

Keshav Memorial Institute of Technology 1


This report includes sample ALPR policies from several jurisdictions to assist readers

in developing their own policies. Readers are also encouraged to review a supplemental

report, “Privacy issues concerning the utilization of automated license plate readers,”

previously prepared by IACP as part of an effort to develop a privacy impact assessment, in

developing ALPR policies for their agencies.

1.1 Background Owners of motorized vehicles driven on public thoroughfares are required by law to annually

register their vehicles with their state bureau or department of motor vehicles, and to attach license

plates that are publicly and legibly displayed. Vehicle license plates generally consist of a series of

alpha numeric characters that reference the license plate to the specific vehicle registered (including

the make, model, year, and vehicle identification number (VIN)) and the owner and/or lien holder of

the vehicle. New York is credited as the first state to enact legislation requiring vehicle registration on

April 25, 1901, with California following in 1902. In Delaware, where numbered license plates were

first issued in porcelain in 1909 beginning with a numbering sequence of 1000, the state changed the

numbering scheme in 1910, beginning with the number “1”, which is reserved for the Governor.

Delaware license plates are sold to the owner of the vehicle and can be passed down generation to

generation. In 2008, a man and his son paid $675,000 in private auction for license plate number

“6”and this figure was matched for Delaware license plate number “11” the following year.

Contemporary license plates, which measure 6 x 12 inches in the United States, feature numbering

schemes that vary from state to state. States typically use numbers or a combination of letters and

numbers in their vehicle license plates. Some states, like Maryland, use stacked letters—one over the

other.

Figure 1.1: Examples of Different State License Plate Numbering Schemes

Connecticut is credited with being the first state to issue vanity plates beginning in

1937, when “motorists with good driving records were allowed to have plates with their

initials (2 or 3 letters).” In Texas any person, non-profit organization, or for-profit entity can

design a specialty license plate for consideration and potential adoption by the state for an

initial deposit of $4,615, which will be refunded to non-profit organizations after 500 of the

plates are sold or renewed. A Texas plate with “PORSCHE” recently sold in private auction

for $7,500, “AMERICA” for $3,000, and “FERRARI” for $15,000.



Fig:1.2:Examples of Specialty License Plates for Passenger Vehicles

Specialty plates generate substantial revenue for states. West Virginia, for example,

reports approximately $1.2 million in revenue from the sale of vanity license plates. Texas is

estimated to have generated approximately $2.1 million in revenue from the sale of specialty

plates through the first 10 months of 2010, Virginia projected potential revenue exceeding $1

million for the sale of specialty plates with company logos, and approximately $600,000 of

revenue generated in Nebraska in 2009 from sales of its “Huskers” license plates. The

California Legislature recently considered a bill to study the potential use of electronic license

plates which would show digital advertisements when the vehicle to which it is attached is

stopped for more than four seconds; the vehicle’s license plate number would display when

the vehicle is in motion.

1.1.1Police use of license plate dataAs noted above, law enforcement practitioners are often searching for vehicles that

have been reported stolen, are suspected of being involved in criminal or terrorist activities,

are owned by persons who are wanted by authorities, have failed to pay parking violations or

maintain current vehicle license registration or insurance, or any of a number of other

legitimate reasons. Victims and witnesses are frequently able to provide police with a

description of a suspect’s vehicle, including in some cases a full or partial reading of their

license plate number. Depending on the seriousness of the incident, officers may receive a list

of vehicles of interest by their agency at the beginning of their shift, or receive radio alerts

throughout the day, providing vehicle descriptions and plate numbers including stolen

vehicles, vehicles registered or associated with wanted individuals or persons of interest,

vehicles attached to an AMBER or missing persons alert, and “be on the lookout” or “BOLO”

alerts. These lists may be sizable depending on the jurisdiction, population size, and criteria

for the list, and can present challenges for the patrol officer.



Officers monitor traffic during patrol, searching for vehicles of interest among their

other duties. When a potential vehicle of interest is observed, the officer will typically

compare characteristics of the observed vehicle and driver with those of the wanted vehicle,

including the license plate number, if known. If warranted, the officer may stop the vehicle to

further investigate. A license plate check will be run on the vehicle, either by the officer using

an in-field computer to initiate the query, or by radioing dispatch for the query. Results of the

query and of the officer’s interaction and investigation of the driver will assist the officer in

determining next steps.

In addition to spotting vehicles of interest, officers on patrol are also alert to vehicles

with expired or missing license plates and annual renewal tags. Failure to maintain current

license plate registration may indicate that one or more of several conditions have not been

met, including failure to secure vehicle insurance, failing compulsory safety and/or emissions

inspections, and simple failure to properly register the vehicle with the state motor vehicle

authority. Some jurisdictions may withhold vehicle registration renewal if the owner has

unpaid parking or traffic tickets. The lack of a plate or current tags may also indicate that the

vehicle has been stolen.

1.1.2 Automated Number Plate Recognition (ANPR)

TechnologyAutomated Number plate recognition (ANPR) technology was invented in 1976 in the

Police Scientific Development Branch (PSDB), Home Office, United Kingdom. The

European Secure Vehicle Alliance (ESVA) notes that the “Provisional Irish Republican Army

(IRA) terrorist bombings in the City of London resulted in the establishment of the ‘ring of

steel’ in 1993 – a surveillance and security cordon using initially CCTV cameras. In 1997,

ANPR cameras, linked to police databases, were fitted at entrances to the ring of steel and

gave feedback to monitoring officers within four seconds.” Implementation continued over

the next several years with forces implementing ANPR systems.

The Home Office Police Standards Unit and the Association of Chief Polices Officers

(ACPO) began testing dedicated “intercept teams” using ANPR across nine police forces in

the multi phased “Project Laser” beginning 2002.The strategic intent of the ANPR strategy

for the Police Services was to “target criminals through their use of the roads.” Intercept

teams, typically ranged in size of 7 – 12 officers and equipped with ANPR, were designed to

enable police to engage criminality on the road and intercept vehicles and drivers wanted in

connection with crime, terrorism, and motor vehicle violations. An analysis of the Laser pilot



projects, which collectively produced over 46,000 arrests, concluded that “ANPR makes a

direct contribution to both national and force objectives and is used on a daily basis to engage

criminals. In comparison to a number of other technology-enabled projects in the criminal

justice area, its success has been remarkable.”

Following success of the Laser pilots, the Police Standards Unit invested £32 million

for development of the National ANPR Data Centre (NADC) and a Back Office Facility

(BOF), which provide data storage and analytic tools for forces in England and Wales, and

support deployment of ANPR at national, regional and local levels. Implementing a single

technology platform in forces across the whole of England and Wales has enabled the UK to

implement universal business practices and technical and data standards. By the end of the

first quarter of 2010, the NADC was receiving approximately 10-12 million license plate

reads per day from over 5,000 ANPR cameras, had the capacity to receive up to 50 million

reads per day, and maintained a database of more than 7 billion vehicle sightings.

ALPR also has many applications beyond law enforcement. It is used by departments

of transportation to monitor travel time on key roadways for better traffic management (where

ALPR captures images of vehicles at two different points on a roadway and calculates travel

times between the two points), automated tolling and toll enforcement, access control, and

congestion charging, among other things.



CHAPTER 2

AN OVERVIEW OF ANPRANPR systems generally consist of a high speed camera with an infrared (“IR”) filter

or two cameras—one high resolution digital camera and one IR camera—to capture images

of license plates; a processor and application capable of performing sophisticated optical

character recognition (OCR) to transform the image of the plate into alphanumeric characters;

application software to compare the transformed license plate characters to databases of

license plates of interest to law enforcement; and a user interface to display the images

captured, the results of the OCR transformation, and an alert capability to notify operators

when a plate matching an agency’s “hot list” is observed. The precise configuration of ALPR

systems varies depending on the manufacturer of the equipment and the specific operational

deployment.

ALPR systems are able to capture up to 1,800 plates per minute at speeds up to 120-

160 miles per hour. Systems range in cost from $10,000 - $22,000, depending on the

manufacturer and the specific configuration specified, and agencies have often been able to

fund acquisition through federal grant funding sources.

2.1Cameras:Camera hardware is significant to the front-end component of any ALPR system.

Since the initial image capture forms a critically important part of the ALPR system and will

often determine the overall performance, ALPR systems typically use still or video cameras

Fig 2.1: Cameras

Specialized for the task. Currently, many of the ALPR systems include a set of high

resolution digital and IR illuminated cameras which allow the ALPR system to capture

images under a variety of light and weather conditions.



2.2 User InterfaceIn vehicle-mounted ALPR systems, captured images are displayed on a user interface

either a dedicated computer for the ALPR system, or use of the in-field computer already

installed in the police vehicle so, the officer can be alerted when a vehicle on one of the hot

lists has been observed in the vicinity of the officer.

The user interface allows the officer to compare the ALPR OCR interpretation of the

license plate number to ensure the accuracy of the read and to see the larger, contextual image

to help the officer in identifying which specific vehicle has the plate of interest. In addition,

the user interface also typically enables the officer to manually enter plates on vehicles of

interest, manage hot list information, deal with alert queues, and run reports.

2.3 Software:

Fig 2.2: operation through computer

As vehicles pass through the field of view of the ALPR camera a picture is taken of

license plate and the vehicle. A series of algorithms are performed on the image to isolate the

plate and render the alphanumeric characters into an electronically readable format. The

sophistication and complexity of each of these algorithms determines the accuracy of the

system. There are six primary algorithms that the software requires for identifying a license

plate:

1. Plate localization – Finding and isolating the plate on the picture.

2. Plate orientation and sizing – Compensates for the skew of the plate and adjusts the

dimensions to the required size.



3. Normalization – Adjusts the brightness and contrast of the image.

4. Character segmentation – Finds the individual characters on the plates.

5. Optical character recognition (OCR) – Translation of images of text into an electronically

readable format.

6. Syntactical/Geometrical analysis – Check characters and positions against state-specific rules to

identify the state of issuance for the license plate.

2.4 Hot lists:Once the OCR read is obtained, the information is then compared against a database

of vehicles of interest, typically known as a “hot list.” Hot list information can come from a

variety of sources, and is discussed in more detail later in this report. The purpose of these

lists is to alert the officer that a vehicle displaying a license plate number that is included on a

hot list has been identified by the ALPR camera. ALPR systems can be deployed in a variety

of ways, including mobile ALPR systems, fixed ALPR systems, and portable ALPR systems.



CHAPTER 3

PROCESS OF AUTOMATIC NUMBER PLATE

RECOGNITIONThe process of Automatic Number Plate Recognition consists of four main stages:

1. Pre-processing

2. License plate localization

3. Character segmentation

4. Character recognition

3.1 Pre-processing:As mentioned before, the system of automatic number plate recognition faces many

challenges. So, this step is essential to enhance the input image and making it more suitable

for the next processing steps. The first step done in the pre-processing is to apply minimum

filter to the image in order to enhance the dark values in the image by increasing their area.

This is mainly done to make the characters and the plate edges bold, and to remove the effect

of the light diagonal strips that appear in the characters and edges of the Egyptian license

plates. This process is followed by increasing saturation of the image to increase the

separation between colours. Then the image is converted to gray scale (taking the luminance

component of NTSC). Then, increasing the image contrast to separate the background from

highlights.

3.2 License Plate Localization:In this stage, the location of the license plate is identified and the output of this stage

will be a sub-image that contains only the license plate. This is done in two main steps.

3.2.1 Locating a large bounding rectangle over the license plateIn this step a rectangle that contains the license plate is located (this rectangle may

also has some extra parts from the four sides), and this rectangle is the input to the next step

for further processing (removing the extra parts, character segmentation then recognition).



Figure 3.1:Large bounding rectangle over the license plate-1

First, Sobel vertical edge detection is applied to the image. Then a threshold of 36 (This value

is determined using trial and error) is applied, such that every edge with magnitude

less than 36 is considered false edge and is set to 0. Then a vertical projection

(projecting on the Y-axis) of the edge detected image is taken and smoothed using an

average filter with width equals 9. It’s obvious that the characters of the plate along

with the plate’s vertical edges will have very strong vertical edges. Moreover, these

edges will sum up horizontally in the vertical projection and a strong peak will appear

in the rows of the plate (These row will be called band). So, the approach is to take

some number of peaks in the vertical projection and processing each of them

individually in the next steps and when a successful band is found, the processing of

the following bands is cancelled. The reason behind taking more than one peak is that

the image may contain objects (logos, road advertisement, etc..) that produce many

vertical edges also these ”false” edges may be centred in the same area so they will

form a peak that may be stronger than the peak of the plate itself.

For each band, we take a sub-image referenced by this band and all subsequent

processing will be applied on this sub-image. Now the problem is to cut the band image from

the left and right to get a bounding rectangle over the license plate (Again, this rectangle

doesn’t have to be tight on the plate). For this sake, a vertical Sobel edge detection is applied

again, but the height is larger than the width of the filter, this is to decrease the effect of false

edges and noise, experimentally, the best size is 6x3 filter .



Figure 3.2: Large bounding rectangle over the license plate-2

Figure 3.3: Large bounding rectangle over the license plate-3

Again a threshold of 30 is applied for the same reason as before. Now, a horizontal

projection of the edge detected band image is taken (projection on the X-axis) and smoothed

using an average filter of large size this time, since there are gaps between the letters and the

projection will have many peaks at the x coordinates where letters exist but it will drop down

in the x coordinates of the gaps. So, smoothing it with average filter of large width will

resolve this problem and many number of peaks will be converted to one wide peak that

represents the range of the X-axis where the plate is located in that specific band we are

working with. The width of the average filter is taken to be the height of the band. Relating

the height of the band with the width of the average filter is very important since over-

smoothing of the projection will merge the plate peak with the other main peaks in the band

like the peak got from vehicle lamps for example (and it already explained why the width

shouldn’t be very small). Now, a predefined number of peaks (It’s already explained why we

take more than one candidate peak not just the strongest one) will be selected from the

smoothed projection.

For each peak, a sub-image is taken according to the range of current peak. So, the

bounding rectangle of the license plate is located. This is will be the input to the next step.

3.2.2 Determining the exact location of the license plate:

Using the sub-image from the last step which contains the license plate with some

extra parts (if any), the following processing is applied to this sub-image. The license plate

may be skewed because of the angle of the camera while image acquisition process. And it is

very important to de-skew the plate to its original orientation, thus making the plate



Figure 3.4: Determining exact location of license plate-1


aligned with the X and Y axes (The reason behind its importance will be clear below). So a

Hough transform is applied to the horizontally edge detected image in order to find the shear

parameters by which the image can be de-skewed to retrieve the standard orientation. After

this operation we have a plate with its axes aligned with the X and Y axes. Then a Gaussian

smoothing filter is applied to smooth the image and remove noise. This operation makes the

characters of the plate bold and increases the characters area along with the effect of

increasing contrast, and subsequently this will ease the process of segmentation and

recognition afterward.

All the above is considered a pre-processing for this step. Next, we aim at finding the

exact band of the plate. In other words, the goal of this step is to cut the top and bottom extra

parts of the previously cut rectangle (but this time the cut will be accurate because we have

limited the area we are working with and moreover we de-skewed the plate). This is done

using the same idea we used previously to get the plate band. It consists of applying Sobel

vertical edge detection, then applying a threshold, then doing a vertical projection (projecting

on the Y-axis), Then getting the strongest peak in this projection and cut the image

accordingly using the range of this peak, thus cutting the exact plate band from the image and

leaving the top and bottom extra parts . This time just the strongest peak is taken since we

already limited the possibility that false edges appear when we cut a rectangle around the

plate and we are sure that the vertical edges produced by the plate’s characters are summed up

correctly in a limited number of rows due to the de-skew operation.




We got rid of the top and bottom extra parts. But we still have extra parts from left

and right that have to be cut to end up with an exact rectangle around the plate. So, a stamp

filter is applied to the sub-image we got from the previous step. This filter is just a blurring

followed by a soft threshold operation. Now the white colour will dominate the plate area .

After this a horizontal projection is done then smoothed using average filter with width equals

40. Then we get the strongest peak from this projection. This peak corresponds to the plate

range on the X-axis. So, a sub-image is cut using the peak range. In many cases when the

colour of the vehicle is bright, the previous operation is not sufficient to cut all the extra

pieces from left and right. So, this is followed by getting Sobel horizontal edge detection,

applying a threshold, then getting the horizontal projection, then smoothing this projection

with average filter of size 40. Then we will get two points that will define range of the peak.

The first point is the point with least x coordinate that has a value (from the smoothed

projection) greater than or equal the average value. The second point is the point with

maximum x coordinate that has a value greater than or equal to the average. We will cut the

image again using these two points we got. And this is the final plate that the next processing

stages will work on.

At the current moment we have a ”candidate” final plate. The next processing stages

are computationally expensive. Also using the fact that all the plates have a very similar (if

not exact) values for some measures like aspect ratio, contrast, average brightness, average

saturation in both the coloured and gray scale plate images. We can begin to reject the plates

based on the previous measures, such that, If we found that the current candidate plate for any

measure has a very far value from the ranges of values for the true plates, It’s simply rejected

and the processing continues on the next candidate plate. But a false plate may pass these

tests, and it will be rejected in subsequent stages. The next stage is to segment characters

from the plate that passed all the measures tests.



3.3 Character Segmentation:This stage is meant for segmentation of the characters from the plate. The output of

this stage is a set of monochrome images for each candidate character in plate.

Figure 3.7: Character Segmentation-1

The first step in this stage is to convert the plate image to a binary image. This is done

using adaptive threshold with a window of size 11 (This is selected using trial and error).

Then a process of noise removal is applied. This is done by getting the connected components

from the binary image based on the 8-neighbourhood using flood fill. For every component,

we decide if it’s a noise or not based on the aspect ratio of the component and based on the

number of pixels in that component. This is based on the fact that the characters of the plate

have a certain range of aspect ratio and a certain range of number of pixels. After removing

the noise components a maximum filter is applied to make the effect of thinning the

characters to make sure that no two components are merged. This is followed by a horizontal

projection, to detect the boundaries between the characters to be able to cut them

individually. The peaks in this projection correspond to the gaps between the characters. So,

we get all of these peaks and a rejection process is applied also, since a true plate has a fixed

range of gaps between characters. So, any plate that has number of peaks that do not fit in

that range, will be rejected. Also, there is a powerful rejection measure; it is the variance of

the characters width (the variance of the spaces between peaks). After this the characters are

cut according to the peaks of the previous projection. Then another set of measures are

computed to reject the false characters that may still exist after the noise removal operation.

These measures are aspect ratio, deviation from average height test, deviation from average

contrast, deviation from average brightness, deviation from hue, deviation from average

saturation. After rejecting false characters, if the number of characters is not located in a

predefined range, then the plate is rejected. Otherwise, the processing is continued.



Figure 3.8: Character Segmentation-2

and for every character a copy of its corresponding location in the gray scale is got. The gray

level histogram is computed for the sub-image of each character, This gray level histogram

will have a standard shape which is one peak at the dark values (this corresponds to the

character’s pixels) and another peak at the bright values (this corresponds to the background)

and some small values between them. So, this gray level image is converted to binary using

the following procedure. First, we find two peaks in the histogram then we find the minimum

value in between, this will be the value of the threshold (thus, every pixel that has a gray level

value less than the mentioned value, will be converted to black, every other value will be

converted to white). This way for converting the gray scale image that contains only a

character to binary one proved to be effective. At this point we have a set of binary images

each contains one character and this is the output of this stage and the input to the next.

3.4 Character recognition:The goal of this stage is to recognize and classify the binary images that contain

characters received from the previous one. After this stage every character must have a label

and an error factor, and this error factor if greater than a predefined value will be used to

reject false characters accidently passed from the previous steps. For the sake of

classification, some features must be collected from the characters. The feature we work with

in this system is the chain code of the contour of the image after dividing it into four tracks

then into four sectors.



Figure 3.9: Character Recognition

Also we used a feed forward artificial neural network trained with back propagation with

sigmoid activation function and the ANN is trained on the chain code feature of the optimal

characters images. The neural network has 4X4X8=128 input neuron, it also has 37 output

neurons corresponds to the Arabic alpha-numeric set of characters except zero, it also ceil

(37+128)/2)=83 hidden neurons.

So, for every character we get the chain code feature and do a feed forward on the

trained FFNN (Feed Forward Neural Network) then the class the corresponds to the neuron

with the maximum value will the predicted class of that character. If the error exceeds a

predefined value then the character is considered a false one and rejected. The plate is known

to have a fixed range of characters that may appear in it, so if the total number of passed

characters does not match this range, then the plate is rejected. Otherwise, the license plate

number is found.



CHAPTER 4

TYPES OF ALPR SYSTEMS

4.1 Mobile ALPR SystemsMobile ALPR systems use vehicle-mounted cameras to capture license plate data and

can be configured in a number of ways to meet specific agency needs. Typically, the

processor is located in the trunk of the vehicle and the data is processed locally to notify the

officer of a possible hit. ALPR cameras are affixed to a vehicle and can be either hardwired

or magnet mounted for a portable (vehicle to vehicle) configuration. They can be integrated

into the light bar, mounted on either the roof or trunk of the vehicle, or within covert housing.



Figure 4.1: Light bar (left) and Covert (right) Mounted Mobile ALPR Cameras

System Portability:Mobile ALPR systems can be hardwired to a vehicle or magnet-mounted, for a

portable (vehicle to vehicle) configuration. Magnet-mounted set-ups offer more flexibility

and allow the agency to relocate the system from one vehicle to another. Consideration

should be given, however, to the location of the hardware and connection cables on the

various vehicle models and the ease with which they can be transferred. Agencies should

weigh the pros and cons of each configuration against the technical and personnel resources

of their agency.

Vehicle space availability:Mobile ALPR system components include cameras, processors, an interface screen,

and keyboard which need to be added to a vehicle. Consideration should be given to the

existing space limitations in both the vehicle cockpit and trunk.

Number of Cameras:Each camera added to the ALPR system on a vehicle provides an additional field of

view and increases the amount of data and images the processor must analyse.

Data Transfer:A variety of methods exist to transfer hot list and ALPR data from the vehicle’s

computer processor. Consideration should be given to whether the agency plans to manually

transfer the hot lists and data files using a USB memory stick or automate the transfer using

wireless or cellular networks.

4.2 Fixed and Portable ALPR Systems ALPR cameras that are permanently affixed to a structure such as a light pole, bridge

or overhead sign.



Figure 4.2: Stationary/Fixed ALPR Cameras on a Bridge and Utility Pole

Fixed and portable ALPR systems require an installation design plan that includes

infrastructure to support the camera system. This infrastructure includes power for the system

and any networking that provides the ability to transmit data between the camera and the

command/information centre.

Some common considerations for fixed systems are:

• Existing physical infrastructure

• Site location

• Available power

• Available network infrastructure

• Number of cameras

• Dispatch requirements

Existing Physical Infrastructure:A great deal of physical infrastructure already exists at key locations along roadways

or potential targets (e.g. sports stadium or power plant). Utilizing established infrastructure

can offer a number of advantages such as reduction in costs associated with setting up a site,

ease of access, and existing power connections. Consideration should be given however to

the agency responsible for the infrastructure as special permits and ongoing maintenance may

be required.

Site location:When choosing site locations for fixed and portable ALPR units, consideration should

be given to whether officers will be routinely stationed nearby and their possible response

times.



Available Power:Fixed and portable systems require power at the location of the camera. The need for

power may limit the possible locations for mounting or require additional resources.

Available network infrastructure:Fixed and portable systems require network connectivity between the ALPR

system’s computer processor (generally located with the camera) and the server receiving

database updates. The updates enable the processors at the camera location to identify

vehicles of interest that have been recently entered into the databases. Agencies should

consider how this network connectivity will be accomplished to ensure successful updates

are received and how the information will be secured.

Dispatch requirements:Fixed and portable systems typically provide alert notifications to the

communications/operations centre. This increases the workload for the dispatch personnel.

Depending on the system configuration, the ALPR system may require an additional

computer screen for the dispatcher to monitor. Dispatch personnel need to be effectively

trained and be able to include the associated actions into their existing responsibilities. It is

also important to ensure that the dispatch facility has sufficient power and space for any

additional computers or servers the ALPR system may require.

Number of cameras: A fixed system typically requires the installation of one camera for each lane of

traffic being monitored. Multiple cameras at one location may improve the ability to locate a

suspect or wanted vehicle.

CHAPTER 5ALPR PERFORMANCE AND POLICIES

A number of factors impact the performance of ALPR systems, and there are several

measures that are relevant to the overall performance of the technology.

1. Capture Efficacy – a measure of the effectiveness of ALPR units to capture the license

plate information of vehicles that pass through the field of view of ALPR cameras. For

example, if 100 cars pass the ALPR unit, what proportion/percentage of vehicles containing

license plates are actually captured (i.e., read) by the ALPR units?



2. Read Accuracy – a measure of the accuracy of ALPR system interpretation of captured

plates with the actual alphanumeric characters of the plate.

3. Matching Effectiveness – a measure of the effectiveness of ALPR units (really of their

underlying software matching algorithms) to accurately match license plates reads to records

contained in the agency’s hot list(s). For example, if the ALPR unit accurately captures or

reads only a portion of a vehicle’s plate, or misreads one or more characters on a vehicle

plate, is the unit (and its software) nevertheless able to match the plate with hotlist records

stored or accessed through the device (perhaps with a scoring factor related to the probability

of an actual match)? This is more a function of the software supporting the ALPR unit, the

calibration of matching algorithms, and a measure of the performance and elasticity of search

parameters.

4. Capture/Read Factors – there are a host of factors that may influence the ability of ALPR

units to capture and accurately read and match license plates. Capture/Read factors include

the following:

a. Character and/or plate colour

b. Plate design factors (logos, stacked characters, etc.)

c. State of origin (i.e., the state which issued the plate)

d. Plate covers or other obstructions (e.g., bent, dirty, trailer hitch obstruction,

etc.)

e. Plate location on the vehicle

f. Interval between vehicles

g. Vehicle speed

h. Lighting conditions (e.g., day vs. night)

i. Weather conditions (e.g., snow, rain, fog)

j. ALPR equipment (e.g., age and/or ability of the ALPR camera)

k. ALPR implementation (e.g., camera angle)

5. Plate design: Each state has multiple license plate designs and plates vary substantially

from state to state. The shape of the characters, amount of contrast between a particular

state’s background and the colour of the license plate characters, and whether the characters

are raised or flat can all impact the accuracy of the OCR read. Some colours, especially

reddish tones, may be difficult for ALPR system OCR software to read.



Figure 5.1: Sample Plate Designs

Poor image resolution:Poor image resolution can result from several factors. License plates can be too far

away for the capabilities of the ALPR camera to capture and motion blur can also occur. Poor

lighting and low contrast due to overexposure, reflection, adverse weather conditions, or

shadows can also result in a poor image quality.

Figure 5.2: Poor Image Quality

Bent, dirty, damaged, or modified plates:

Because many ALPR systems use reflectivity and the contrast created by the

alphanumeric characters, plates that are bent, dirty, damaged, or modified may cause the

ALPR software to misidentify a character.

Figure 5.3: Bent, Dirty, or Damaged Plates

Plate location:Occasionally, an object might obscure all or a portion of the license plate and interfere

with accurate OCR. Oftentimes the object is a tow bar, dirt on the license plate, or a loaded



bike rack; other times the object may be a ALPR circumvention device. The National

Policing Improvement Agency (NPIA) and the Association of Chief Police Officers (ACPO)

have specified minimum performance capabilities of ALPR technologies in capturing and

reading license plate information for UK and Schengen Community plates.

Figure 5.4: Obstructed Plates

Table 5.1: ‘Capture’ and ‘Read’ rates for All United Kingdom (UK)

Type of System Capture RateCorrect Read

Rate

Overall capture &

correct read rate

Static ANPR Camera 98% 95% 93.1%

CCTV Integrated ANPR

(Dual purpose CCTV and

ANPR Camera)

85% 85% 72.0%

Mobile ANPR Camera

(Stationary)98% 95% 93.1%

Mobile ANPR Camera

(Moving)80% 85% 68.0%

Given the fact that in static ALPR implementations the camera capturing the license plate is

stationary, whereas mobile implementations involve mobile cameras and potentially mobile

target vehicles, variations in capture efficacy and read accuracy rates are expected and

observed. Slightly lower performance rates are acceptable for capturing and reading plates of

just Schengen member countries.

Table 5.2: ‘Capture’ and ‘Read’rates of Schengen Community

Type of System Capture RateCorrect Read

Rate

Overall capture

& correct read

rate



Static ANPR Camera 85% 80% 68.0%

CCTV Integrated

ANPR

(Dual purpose CCTV

and

ANPR Camera)

85% 80% 68.0%

Mobile ANPR Camera

(Stationary)85% 80% 68.0%

Mobile ANPR Camera

(Moving)75% 80% 60.0%

Law enforcement agencies in the United States must deal with license plates from other

states, as well as international jurisdictions, which can pose difficulties if the units are not

configured to “read” plates from multiple jurisdictions. ALPR manufacturers are constantly

upgrading their equipment and software to address this issue. Minimum performance

standards for ALPR in the United States are still at an early stage. IACP is managing an NIJ-

funded project to develop technical performance standards for ALPR systems. The goals of

the project are to a) identify the specific performance parameters that are critical to ALPR

functions, b) develop metrics to accurately measure their performance, and c) establish

protocols for the testing of the equipment by an independent laboratory.

5.2 ALPR Policies:In order to be effective, ALPR technology must be properly implemented and

integrated into the daily operations of law enforcement agencies. Developing and enforcing

policies defining the strategic objectives of an agency’s program, training requirements,

deployment options, operating procedures, hot list management, proper use and maintenance

of the technology, and data collection, retention, sharing, and access enables law enforcement

to effectively manage ALPR.

Agency policies typically:

1. Define the ALPR system and its data as “for official use only” (FOUO),

2. Restrict and audit queries of the ALPR dataset, and



3. Require that all operators of the equipment receive proper training before use.

Essential components of that training include:

1. Ensuring that appropriate hot list information is as current and accurate as possible, and

2. Clear directions that when the ALPR unit matches an observed vehicle’s license plate with

a record in the hot list and alerts the officer (also known as a “hit”), that the officer must

verify

a. that the ALPR “read” was accurate (i.e., that the ALPR OCR software has

properly interpreted the license plate number),

b. that the issuing State matches the plate on record, and

c. that the circumstance that triggered the alert is still current, e.g., that the

vehicle is still wanted or stolen.

3. If the record in the hot list was created based, not on the status of the vehicle, but rather on

the status of the registered owner (e.g., the owner has an outstanding warrant for arrest, or has

had their driving privileges suspended or revoked), the officer must also be cognizant of the

fact that the driver may not be the registered owner.

Additionally, depending on the nature of the alert (e.g., a “hit” on the Terrorist Watch List),

the officer may be directed to notify another agency (e.g., the Terrorist Screening Centre) and

hold the person, surreptitiously watch but not contact the person, or simply document the

contact and forward the information to others.

Nearly half of responding agencies (19 agencies, 48%) indicated that they had a policy

addressing ALPR use and operations, and six agencies (15%) noted they were in the process

of developing or planning one. Among agencies that have or are developing ALPR policies,

the policies usually address data access (68%), data retention (48%), and data sharing (44%).

Table 5.3: Policy Issues Addressed by Agencies

Policy Issues n %

Data access 17 68%

Data retention 12 48%

Data sharing 11 44%



Deployment 6 24%

Data quality 4 16%

Other 4 16%

CHAPTER 6

ADVANTAGES OF USING ANPR TECHNOLOGY



With the help of this technology, pre-paid parking members can be easily

differentiated from non-members. With the ability to capture the driver’s image, car hijacking

can be prevented to a large extent. By having a list of stolen cars and unpaid fines in the

database, identification of cars with such history can be easily made and real time alerts can

be obtained, as soon as the car is located on a highway or at a signal. Traffic congestions can

be lowered, by diverting different vehicles into different lanes with the help of their entry

permits. Those persons or vehicles whose presence is barred from entry can be easily

identified.

License Plate Recognition (LPR) technology uses image processing to identify vehicle

license plates. The technology is rapid and able to identify and record a license plate number

under most ordinary driving conditions, including when a car is moving at high speed and at

night. It offers advantages in law enforcement, security and vehicle access.

Traffic EnforcementLicense plate recognition technology is an advantage in several areas of traffic

enforcement. For example, a license plate recognition device can be put at an intersection and

record the license plate of a car that runs a red light. A citation can then be sent to the

registered owner of the car. Similarly, an LPR device can be placed in an area where speeding

is a common problem, and record the infraction for follow-through with a citation.

Law EnforcementLicense plate recognition technology has several applications in areas of law

enforcement. An LPR device can be mounted on a patrol car to record plates of passing cars.

This information can then be compared with the law enforcement data base for vehicles

associated with a crime. This is useful in Amber alerts, finding stolen vehicles and executing

felony warrants.

Vehicle AccessAn LPR system is also useful for vehicle access. On toll roads, these systems can be

used to allow cars to pass through toll gates without stopping. The LPR records the plate of

the car and associates the number with the registered owner. The bill can then be sent by mail.

This technology also can be used in secure, gated locations. When a vehicle recorded in a

database approaches a security gate, the system recognizes the license plate, and the gate

opens automatically. This allows the driver to pass without interruption and eliminates the

possibility of an entrance code being stolen.





DIFFICUTIES AND FUTURE SCOPEThere are a number of possible difficulties that the software must be able to cope with.

These include:

Poor file resolution, usually because the plate is too far away but sometimes resulting

from the use of a low-quality camera.

Blurry images, particularly motion blur

Poor lighting and low contrast due to overexposure and reflection or

shadows.

An object obscuring (part of) the plate, quite often a tow bar, or dirt on the plate.

Read license plates that are different at the front and the back because of towed

trailers, campers, etc.

Vehicle lane changes in the camera's angle of view during license plate reading.

A different font, popular for vanity plates (some countries do not allow such plates,

eliminating the problem).

Circumvention techniques

Lack of coordination between countries or states. Two cars from different countries

or states can have the same number but different design of the plate.

While some of these problems can be corrected within the software, it is primarily left

to the hardware side of the system to work out solutions to these difficulties. Increasing the

height of the camera may avoid problems with objects (such as other vehicles) obscuring the

plate but introduces and increases other problems, such as the adjusting for the increased

skew of the plate.

On some cars, tow bars may obscure one or two characters of the license plate. Bikes

on bike racks can also obscure the number plate, though in some countries and jurisdictions,

such as Victoria, Australia, "bike plates" are supposed to befitted. Some small-scale systems

allow for some errors in the license plate. When used for giving specific vehicles access to a

barricaded area, the decision may be made to have an acceptable error rate of one character.

This is because the likelihood of an unauthorized car having such a similar license plate is

seen as quite small. However, this level of inaccuracy would not be acceptable in most

applications of an ANPR system.

Future scope:



In recent years, digitisation and IP-based communication networks have allowed

Automatic Number Plate Recognition (ANPR) to achieve ever-greater utility.

Digitisation and ANPR have grown alongside each other and it has been natural for

clients to include ANPR when upgrading, mainly because by the time digitisation reached

traffic enforcement clients had a more open mind towards computers and expected their

enforcement units to do what their home PC was able to. From the manufacturer's

perspective, our aim is to make the operational aspects of enforcement as easy as possible and

adding ANPR software to our digital units was a natural step.

25-30 percent growth:The ANPR market will grow by around 25-30 per cent in the next five years.

Enforcement solutions combining, for example, speed measurement, camera and ANPR will

drive future growth, according to ErnoSzucs of ARH, Inc. Electronic identification of

vehicles using DSRC onboard units will always require some complementary, non-intrusive

technique such as ANPR. The reason is obvious: assuring road administrations that every

vehicle can be identified electronically would require complete agreement between all

countries and vehicle manufacturers. That's far from being possible at the moment.

CHAPTER 8



APPLICATIONS OF ANPRAutomatic Number Plate Recognition has a wide range of applications since the

license number is the primary, most widely accepted, human readable, mandatory identifier

of motor vehicles. ANPR provides automated access of the content of the number plate for

computer systems managing databases and processing information of vehicle movements.

Below we indicated some of the major applications, without the demand of completeness.

Parking:One of the main applications of ANPR is parking automation and parking security:

ticketless parking fee management, parking access automation, vehicle location guidance,

car theft prevention, "lost ticket" fraud, fraud by changing tickets, simplified, partially or

fully automated payment process, among many others.

Access Control:Access control in general is a mechanism for limiting access to areas and resources

based on users' identities and their membership in various predefined groups. Access to

limited zones, however, may also be managed based on the accessing vehicles alone, or

together with personal identity. License plate recognition brings automation of vehicle access

control management, providing increased security, car pool management for logistics,

security guide assistance, event logging, event management, keeping access diary,

possibilities for analysis and data mining.

Motorway Road Tolling:Road Tolling means, that motorists pay directly for the usage of particular segment of

road infrastructures. Tolls are a common way of funding the improvements of highways,

motorways, roads and bridges: tolls are fees for services. Efficient road tolling increases the

level of related road services by reducing travel time overhead, congestion and improve

roadways quality. Also, efficient road tolling reduces fraud related to non-payment, makes

charging effective, reduces required manpower to process events of exceptions. License plate

recognition is mostly used as a very efficient enforcement tool, while there are road tolling

systems based solely on license plate recognition too.

Border Control:Border Control is an established state-coordinated effort to achieve operational

control of the country's state border with the priority mission of supporting the homeland's

security against terrorism, illegal cross border traffic, smuggling and criminal activities.



Efficient border control significantly decreases the rate of violent crime and increases the

society's security. Automatic number plate recognition adds significant value by event

logging, establishing investigate-able databases of border crossings, alarming on suspicious

passing’s, at many more.

Journey Time Measurement:

Journey Time Measurement is a very efficient and widely usable method of

understanding traffic, detecting conspicuous situations and events, etc. A computer vision

based system has its well known downfalls in Journey Time Measurement, while Automatic

Number Plate Recognition has provided its viability: vehicle journey times can be measured

reliably by automatic number plate recognition-based systems. Data collected by license plate

recognition systems can be used in many ways after processing: feeding back information to

road users to increase traffic security, helping efficient law enforcement, optimising traffic

routes, reducing costs and time, etc.

Law Enforcement:Automatic number plate recognition is an ideal technology to be used for law

enforcement purposes. It is able to automatically identify stolen cars based on the up-to date

blacklist. Other very common law enforcement applications are red-light enforcement and

over speed charging and bus lane control.



CHAPTER 9

CONCLUSIONThe objective of this paper was to study and resolve algorithmic and mathematical

aspects of the automatic number plate recognition systems, such as problematic of machine

vision, pattern recognition, OCR and neural networks. The problematic has been divided into

several chapters, according to a logical sequence of the individual recognition steps. Even

though there is a strong succession of algorithms applied during the recognition process,

chapters can be studied independently.

Law enforcement agencies throughout the nation are increasingly adopting automated

license plate recognition (ALPR) technologies to enhance their enforcement and investigative

capabilities, expand their collection of relevant data, and expedite the tedious and time

consuming process of manually comparing vehicle license plates with lists of stolen, wanted,

and other vehicles of interest. Police officers, sheriff’s deputies, and other law enforcement

practitioners are often on the lookout for vehicles that have been reported stolen, are wanted

in connection with a crime or traffic violation, are suspected of being involved in criminal or

terrorist activities, are parking violation scofflaws, have failed to maintain current registration

or to comply with statutory insurance requirements, or any of a number of other legitimate

reasons.

ALPR systems function to automatically capture an image of the vehicle’s license

plate, transform that image into alphanumeric characters using optical character recognition

or similar software, compare the plate number acquired to one or more databases of vehicles

of interest to law enforcement and other agencies, and to alert the officer when a vehicle of

interest has been observed.

ANPR solution has been tested on static snapshots of vehicles, which has been

divided into several sets according to difficultness. Sets of blurry and skewed snapshots give

worse recognition rates than a set of snapshots which has been captured clearly. The

objective of the tests was not to find a one hundred percent recognizable set of snapshots, but

to test the invariance of the algorithms on random snapshots systematically classified to the

sets according to their properties.



BIBLIOGRAPHY [1] Shan Du ; IntelliView Technol., Inc., Calgary, AB, Canada ; Ibrahim, M. ; Shehata, M. ;

Badawy, Wael; Automatic License Plate Recognition (ALPR): A State-of-the-Art Review".

IEEE. 1 Feb 2013. Retrieved 2014-01-09.

[2] C.N.E. Anagnostopoulos, I.E. Anagnostopoulos, I.D. Psoroulas, V. Loumos and E.

Kayafas, License Plate Recognition From Still Images and Video Sequences: A Survey, IEEE

Transactions on Intelligent Transportation Systems 9 (2008), no. 3, 377–391.

[3] C.N.E. Anagnostopoulos, I.E. Anagnostopoulos, V. Loumos and E. Kayafas, A License

PlateRecognition Algorithm for Intelligent Transportation System Applications, IEEE

Transactions on Intelligent Transportation Systems 7 (2006), no. 3, 377–392.

[4] S.-L. Chang; L.-S. Chen; Y.-C. Chung; S.-W. Chen, Automatic License Plate

Recognition, IEEE Transactions on Intelligent Transportation Systems 5 (2004), no. 1, 42–53.

[5] S. Draghici, A neural network based artificial vision system for licence plate recognition,

International Journal of Neural Systems 8 (1997), no. 1, 113–126.

[6] B. Enyedi, L. Konyha, C. Szombathy and K. Fazekas, Strategies for fast license plate

number localization, Proceedings of the 46th International Symposium Electronics in Marine,

Elmar 2004, Zadar, Croatia, June 16-18, 2004, IEEE Press (2004), 579–584 .

[7] J.-W. Hsieh, S.-H. Yu and S.-H. Yu, Morphology-based License Plate Detection from

Complex Scenes, IEEE Proceedings of the 16th International Conference on Pattern

Recognition, Qubec City, Canada, August 11-15, 2002, Vol. 3 (2002), 176-179.

[8] O. Martinsky, Algorithmic And Mathematical Principles Of Automatic Number Plate

Recognition Systems, B.Sc. thesis, Department of Intelligent Systems, Faculty of Information

Technology, Brno University of Technology (2007).

[9] J.R. Parker and P. Federl, An Approach To Licence Plate Recognition, University of

Calgary 1996.

[10] H. Sarukhanyan, S. Alaverdyan, and G. Petrosyan, Automatic Number Plate Recognition

System, Proceedings of the 7th International Conference on Computer Science and

Information Technologies, Yerevan, Armenia, September 28-2 October, 2009, Electronic

Copy of the CSIT 2009 Proceedings, 347-350.


Date post:	08-Jan-2017
Category:	Engineering
Upload:	neeraja-gandla
View:	263 times
Download:	0 times

Automatic Number Plate Recognition

Engineering