ALPR

AUTOMATIC LICENSE PLATE

RECOGNITION (ALPR)

M G Niketh , 08ec33

INTRODUCTION

Reading a license plate is the first step in determining the

identities of parties involved in traffic incidents. An efficient

automatic license plate recognition process may become the core

of fully

computerized road traffic monitoring systems

electronic fee collection solutions

surveillance devices

safety supervision systems.

It is important that the recognition accuracy of such a process is

very high. Tracking and registering dangerous behavior in traffic

may be used for prosecuting offenders.

ASSUMPTIONS

Input is an image of a stationary Car.

Only the most common type of license plates (single line) will be

dealt with.

The license Plate has a White background with text written in

Black.

PROCEDURE

License-Plate Recognition System consists of following main

modules:

License plate detection,

Character segmentation and

Cleaning phase

Classification phase

LICENSE PLATE EXTRACTION

The license plate region is characterised by a row of

transitions from dark to light and vice versa (we call such a

transition an edge).

Algorithm

First Convert from RGB to grayscale

Generate edge image corresponding to grayscale image

( Sobel edge detection )

Perform row profiling to determine the Y coordinates of

license plate region .

Segment the edge image along this (y1-y2 region)

Perform column profiling to determine the X coordinates of

license plate region .

Segment the grayscale images along the X and Y set of

coordinates so obtained above.

Row profiling

Y region

Column profiling

X region

More edges here

SEGMENTATION PHASE

Column that contains part of a

character is darker than a column

that contains the background of the

license plate .

Algorithm

Perform Column profiling .

( Multiple broad peaks obtained

for different characters )

Segment the images along the

lines between the peaks , to

obtain character bitmaps .

CLEANING PHASE

Removal of all dark pixels from the character bitmap that do not belong to the characters.

Eg. Black bars above and below the characters ,dirt spots, or nuts and bolts.

Algorithm

Convert from grayscale to binary

Remove all rows and columns containing only white or only Black pixels , or say (99 % white, black pixels )

Remove all small groups of black pixels .

The cleaned character bitmaps. All dark pixels that are not part

of the character are removed

CLASSIFICATION PHASE (FEATURE DETECTION)

Determine the character type from the character

bitmap obtained based on the unique features

each character possesses .

Algorithm

Perform the row profiling to obtain ‘row peaks’

Perform the column profiling to obtain ‘column

peaks’.

Find number of junctions of the character.

Find number of end points of the character.

2 column peaks

+ 3 row peaks

The character

could be any of

{5,6,8,9,B,G}

1 End point

+ 1 junction Row(end point) is higher than

row(junction). Thus character is ‘6’

TEMPLATE MATCHING

Alternative solution to the classification phase is –

Algorithm

The character bitmap is contrast enhanced .

All the character bitmaps are normalised and fit into a

box of the size of the standard templates .

Find cross correlation (measure of similarity) between

the set of templates and the obtained character bitmap

The character is matched to the one with the highest

CF value .

TEMPLATE

DATABASE

DRAWBACKS

DRAWBACKS

The algorithm so discussed fails to properly

differentiate between (B,8) (0,D) (2,Z).

The algorithm assumes that there are not more

than 1 candidate for license plate region .

SOLUTION:

Use the Aspect ratio of the standard license plates to

obtain the right candidate among many )

Compare the edge density (number of edges per unit

area) of the various candidates .The highest one is the most

appropriate one .

CONCLUSION

What is trivial for the human eye may appear a very difficult task

for the computer, but still computer vision is very powerful tool

that provides us the capabilities to perform very useful operations

as the one we implemented in this project.

Application

•Parking lot management

•Automatic toll collection enforcement

•Traffic enforcement statistics

•Border surveillance

•Stolen vehicle search