Security Sales & Integration January 2022 - 48

Video
Adopting ALPR
With edge-based ALPR, all of the image
processing is done in the camera at the edge.
This means that the camera transmits a drastically
reduced amount of data over the network,
namely the license plate read, a compressed
snapshot of the vehicle, and associated metadata,
including date, time, and location. This
requires much less bandwidth and expense than
transmitting live video feeds.
Machine learning is making ALPR more attractive
to end users and easier for SIs to integrate into
their portfolios.
In addition, because the processing is
done in the device, even if the network goes
down, the camera continues to read images
and collect data. Then, it can send all the
stored information when communication
has been re-established.
How Machine Learning Helps
Machine learning is making ALPR more attractive
to end users and easier for SIs to inteALPR
cameras can read
features from video images like
color of a car and its make. If a
witness to an accident cannot
remember a car's license plate
but knows it was a pickup truck
or red Toyota, investigators can
use that info as search criteria
and the system can pull up
each instance from its feeds.
grate into their portfolios.
In the past, image processing was done
through rules-based algorithms. Data scientists
wrote the code that enabled the computer
to determine if a character was an 'X' or a 'Y.'
Sometimes, a change in the design of a plate's
background could interfere with the read. In
the state of Oregon, for example, a Crater Lake
plate with a mountain range in the background
was developed. Using rules-based algorithms,
some ALPR systems read the 'I' as a 'T'. And
this is not an isolated problem.
Many states in the U.S. have 80-90 different
license plate designs. You can have one that
indicates your alma mater, your profession, or
your favorite football team. Features on these
plates can significantly affect readings, and, as
a result, data scientists have had to write more
code for the rules-based algorithms to account
for all the changes in pixels around the letters
and numbers on each plate.
Now, with machine learning, data scientists
simply write the code that allows a computer to
learn to read plates. During supervised machine
learning, the camera captures images and then a
trainer labels these images, for example indicating
that this is a letter 'I.' When the camera has
ingested enough labeled images, the algorithm
allows the computer to look at an image and
make a mathematical guess as to what the letter
is. In the case of the Crater Lake license plate,
the computer could be trained to recognize that
the mountain range in the background is not
part of the letter 'I.'
By moving to machine learning, ALPR
has gone beyond what rules-based algorithms
could identify. Today, ALPR cameras can read
other features from video images, like the color
of a car and its make. This means that, if
a witness to an accident cannot remember a
car's license plate but does know that it was a
pickup truck or red Toyota, investigators can
use these descriptions as search criteria and the
system will be able to pull up each instance
from its feeds.
Who Is Using ALPR Technology?
Law enforcement and parking were the first
verticals to adopt ALPR cameras and systems.
The police started using ALPR technology in
patrol cars to read license plates and compare
them with a hotlist of wanted or stolen cars.
Parking enforcement then adopted the technology
to help manage permits and identify parking
violations.
But ALPR is now being used in other verticals.
This is, in part, because machine learning
has increased the read accuracy of purpose-built
cameras to the same level as other technologies,
including RFID. Government organizations
are now expanding their use of ALPR to include
different applications, like tolling. In addition,
the private industry is starting to adopt
the technology for a variety of uses, including
48
Security Sales & Integration JANUARY 2022
securitysales.com
LAWR/ADOBE.STOCK.COM
http://www.securitysales.com
Security Sales & Integration January 2022

Table of Contents for the Digital Edition of Security Sales & Integration January 2022
