Take computer vision a branch of
AI, which itself is a branch of computer science with cv. Neil Haboush can
create a system that does a subset of things that the human visual system can
do in CV. The system can analyze a picture taken by a camera to understand,
what's in the picture.
For example it can recognize
objects like cars, streetlights and people. While humans have no problems with
these kinds of things. It's difficult for computers to do the same thing
because all it sees is a rectangular array of pixels. Each with a certain
intensity and color we need the computer to somehow take a step back and look
at the pixels. Around the neighborhood of a point instead a popular way to do
that is to slide a window across the image and at each point look for
higher-level features.
If you can somehow summarize
what's in each window, then you can run the same process. Through summaries and
repeat the process to hope to recognize objects computers, do this through a
network of nodes called neural networks. An image can be fed into the network
and the concept of sliding one window or the image is called convolution. So
the network itself is called a convolutional network or CNN. So that's Artificial Intelligence Park now let's look at the
business part.
Say you have a business that
manages a parking lot in the city of Montreal. We have three types of customer’s
members who pay a monthly fee and whose credit card information. We already
have members who pay per use and whose credit card information. We already have
non-members who just pay by usage and whose credit card information we do not
have. We want to use a computer vision based license-plate recognition system
to simplify operations. When cars come in or go out their plates are
automatically red and the members are charged appropriately. Others are charged
at exit but there is no concept of having to take a ticket as customers. Drive
in a customer simply pulls up into the parking lot and parks on entry. The CD
system registers her license plate and logs her check-in time when she leaves.
If she's a member, she simply drives
out, if she's not then the gate does not open until she pays at the credit card
machine. Here's what the process looks like a non-member may choose to become a
member with a single acknowledgment. Since
we already have all the required information for membership like the credit
card license plates and car model to make this all happen. The license-plate
recognition system the registration system and the billing system. All have to
be integrated there are a few other processes and systems as well but as you
can see the CV system is simply one part of the end-to-end process. Now let's
look at some exception conditions, what happens if the car comes in with dirty
license plates and the system is not able to read it.
We have a few options here one is
to send the car through a manual process where someone reads off the license
plate and enters. It manually this will
increase operational costs that includes an employee and perhaps construction
of a separate lane for such processing. Another is to write it off because the
number of such occurrences, may be so small that it's not worth paying somebody
to handle. Exceptions this decision needs to be made based on historical data,
because you don't want to write off too much revenue another option is to use
two systems want to read the front of the car and another to read the license
plate at the back of the car.
This will reduce a number of exceptions
but some states do not enquire front and back license plates. Yet another
option is to install a water spray module to wash off the soil in the snow as
the car enters. This will increase the reading accuracy and is similar to
cleaning the data before you feed it into a CV system or a machine learning
system. The point is that as you look at the total implementation of the process
you will see that artificial intelligence plays only a small part you have.
To make many other decisions
about customer experience, speed cost employees processes etc to make the
solution reality. So don't get an armoured by the AI technology. We might even
have to throw in other technologies like IOT and blockchain to make these
processes even more robust.
If you like neil haboush articles please comment below.
0 comments:
Post a Comment