Going My Way
Expectations for AI go far beyond what you might think
- By Thomas Cook
- Feb 01, 2020
Artificial Intelligence (AI) continues to be a hot
topic in the security industry. While true AI (a
computer’s ability to think and act like a human)
is still decades away, subsets of AI such as
deep learning and machine learning have enabled
computers to make significant strides, enabling machines to examine
large amounts of data to provide deeper insights. So, while
we do not find ourselves in the realm of true artificial intelligence,
using the term AI to describe any aspect of it, seems here to stay.
As more “AI-based” cameras come to market, it is interesting
to hear how end users think about deploying the technology.
Recently, I participated in an end-user conference attended by a
diverse group of Fortune 500 organizations along with government
and transportation groups.
We asked them what they wanted from AI-based cameras in
the future and the expectations expressed were far beyond what
you might expect. The promise of AI has infiltrated almost every
corner of our business and personal lives. The prevailing wisdom is
that AI has the potential to do anything, so in addition to advancing
their security capabilities, customers are looking to AI for help
in solving their operations and business challenges as well.
This trend is bolstered by consumer products and services that
continue to push the envelope too, setting further expectations
for AI for products we use in our day-to-day lives. For example,
if you buy a new car today, there are sensors everywhere which
can inform you if an obstacle is approaching as you reverse, slow
you down automatically when you come up behind another car
or subtly push you back in the lane when you drift out.
Autonomous vehicles have put AI front and center as they use
this technology to enable cars to drive themselves. These ideas of
computer assistance are becoming a common expression of AI
across all of our devices.
It is a challenging part of the future, whether we like it or not,
and security systems will no doubt be following the same path. It
is easy to imagine smart systems making decisions with little to
no human intervention—automating reports and sending them
to police stations, end users, directors and IT departments without
anyone having to directly interact with it.
Because AI is marketed everywhere, it is not surprising that end
users expect much more than what the industry currently offers
them. While discussing the fact that AI-based cameras can recognize
the color of a person’s shirt and pants, one of our customers
said, “Yes, that’s useful, however, we would like to be able to detect
our company logo on shirts. When an employee enters a camera’s
field of view, we want to know if they are wearing the appropriate
company attire and are therefore compliant with standards.”
Even before customers receive their first AI-based cameras,
they are evolving their expectations of the technology to fit their
needs. Another example came from a food processing customer
that wanted their cameras to read labels on packages of meat to determine if they are nearing the expiration date.
Today, while we might be able to identify a piece of chicken,
we currently have no way to tell if it is expiring. Perhaps we can
partner with another sensor company to scan packaged meat
items when they pass through a distribution center and satisfy
the customer’s request, but you can see where this is going. Likewise,
a major pizza company asked if they could see if the pizza is
placed correctly in the box before it is shut to make sure the toppings
stay on the pizza rather than sticking to the box. Obviously,
it is a part of their process they want to improve.
Opportunities for Business and
Operations Intelligence Are Growing
The security industry, which until recently was seen as a commodity,
now has the opportunity to reinvent itself as cameras
and supporting infrastructure also become smart devices. There
is plenty being done on the mechanical side, but also on the electronic
video side. People recognize the advancements and the innovation
possible at the camera level as the devices become increasingly
Having more intelligence at the edge gives end users a dramatic
advantage to process data and makes changes in real time.
In the past, we would have to take the data and pass it on to servers
at the head end, or in NVRs, but we are getting to the point
now where the valuable data we mine can be shared from cameras
directly to additional systems and services allowing business to
do more with it.
Partners can aggregate the data into charts and graphs while
linking into POS systems as well as intrusion and facial recognition
systems. We can take all of this valuable information from
the edge and then disseminate it where it is needed.
Our most recent end-user conference gave us a tremendous
amount of feedback, but while the sky may be the limit in the customer’s
mind, it is important to set expectations accordingly when
it comes to uses of AI. In some conversations, it was clear that end
users are assuming they can teach AI-based systems themselves
without realizing just how complex an endeavor that can be.
For a deep-learning algorithm to correctly recognize an object,
it has to have been shown that object hundreds of thousands
of times (if not much more) in varying environmental conditions.
When the algorithm gets something wrong, it has to be corrected
and tested again. While this technology might be here before we
know it, machine learning and deep learning are best left to experts
in the technology, and if we want to keep things 100 percent
accurate, it is important for people to know that training an algorithm
is a long way from teaching your smart speaker or phone to
recognize your voice.
The Age of Analytics
Traditional motion analytics are pixel-based, and depending on
the environment in which they are used, can easily generate false
positives. For example, if a bag is left in an airport, and an operator
is doing forensic research to find out when the bag was moved,
a pixel-based motion analytic cannot discern the difference between
the item physically being removed or if a person is standing
in front of the bag and blocking it from the camera’s view.
So, while these analytics are useful, there are many instances
where false-positive data is likely to occur. Likewise, real-time
alerts from video security systems provide an excellent line of defense,
but the susceptibility to generate false alarms in the past,
such as wind blowing trees, have caused many organizations to
pull back on all but the most guaranteed scenarios.
False-positive alerts can be costly for a company. We have
lived with technology in the previous decade that was not 100
percent accurate, and instead of losing credibility within the organization,
many security teams have opted for the safer route
which was to avoid false alarms at all costs. The challenge now
is to demonstrate to end users and the entire industry that the
technology has improved so significantly, thanks to deep learning,
that it is reliable and can be trusted.
We are going to see more AI-assisted products coming out in
2020. Cameras with deep-learning technology can recognize programmed
objects and describe unique characteristics such as color
and worn accessories. They can identify a person with or without
glasses. They can detect whether you are holding a cell phone to
your ear. This represents an exponential improvement over pixelbased
motion analytics in their ability to prevent false alarms.
It is All About the Data
AI is gaining momentum every day, partly because it is poised
to infiltrate every aspect of business and our personal lives. Ultimately,
it is all about organizing and making sense of data. Modern
cameras can gather much more than video image data. They
can also classify sounds, recognize and count objects, display heat
maps and know when tampering occurs.
They are an important data gathering tool that is already well
accepted and commonplace in our world. This continues to move
them well beyond their original commodity status and into a
revenue-generating device that can provide multiple services for
business intelligence applications as well as operations and process
measuring and metrics.
The key takeaway here is that customers, when asked what
they want to see from AI technology, did not reply with security
workflow improvements (possibly because they do not know
what is achievable yet). Instead, many replied with operations
and processing needs. However, there is a clear desire for smart
cameras to assist more in business operations.
Let me give you one final example: A large airport expressed
interest in having cameras installed at every gate so that staff
could be alerted when an aircraft arrives at the gate. Currently,
the only way they confirm this is to have a person physically look
out the window.
This ties back to security too, of course, knowing whether
the door should be opened or not. AI-based
technology is certainly going to bring exciting
change and plenty of challenges to our industry
in the years ahead.
This article originally appeared in the January / February 2020 issue of Security Today.