The Next Evolution
Video security posed for revolutionary technology to enter
- By Shawn Guan
- Mar 01, 2018
It is time to pay attention to Artificial Intelligence (AI). With
Apple Siri, Google Assistant and now Amazon Echo, voice recognition
AI interacts with users worldwide thousands of times a
day. In visual recognition, AI is embedded in every day consumer
products like Facebook’s DeepFace and Apple’s Face ID. Future
AI projects in development include self-driving cars, autonomous
drones and robots, and Google computer program AlphaGo learning
AI. This revolution is taking every industry in the global economy
by storm and video security is no different.
There are two ways AI is changing and continues to influence the
industry: improving efficiency and safety at lower operating costs
and unlocking brand new capabilities never before thought possible,
capabilities that would have needed a massive team of humans to
manually do it—think Big Data. Many companies may focus on the
former because it saves money, yet the most forward-thinking ones
should instead seek for the latter because it creates growth.
Current Landscape of the
Video Security Industry
The video security industry is currently estimated to be about a $16
billion market1 and is projected to grow to $90 billion by 2020 when
adjacent fields such as alarm monitoring and access control are accounted
for.
The video security industry’s secular growth is driven by a need
for greater security in public areas, as well as Smart City initiatives
that feature many networked objects and sensors communicating
with each other known as Internet-of-Things (IoT). The result has
been the introduction of 66 million cameras coming online every year
and more than seven billion hours of footage being recorded around
the world.
The problem sprouting from this situation is simple: There is too
much data, and too few eyes watching it. A research project by the
U.K. Police Scientific and Development Branch found that the more screens a single person watches, the less effective they become. In fast
moving environments like casinos, a person can top out at just five
screens. The ideal situation is one screen to one person, but no center
can afford to run this way.
The result is that video becomes mere evidence collection with
little or no actual security value. We see the cameras at schools, public
transits, offices or residences but do not feel any more secure when we
do. That’s where AI can create value.
Introduction of AI into
Video Security Systems
AI is a broad field that encompasses a variety of different algorithms
and approaches to replicate human intelligence, autonomy and proactivity.
These approaches have not had success in even closely approximating
human performance yet. However, in recent years, what
has been driving the greatest innovation in the AI field has been an
approach called machine learning. Made possible by the development
of powerful graphics cards and the cloud, these smart and IoT devices
can analyze large amounts of data to drive insights and action.
By having a computer “look at” many examples of something, a
model can be developed that would be able to identify this specific
behavior. These approaches have been found to be very effective in
computer vision tasks like handwriting recognition. Researchers and
industry executives think video security can be next.
Opportunities and Challenges
Many companies will see AI as a cost-saving measure that can simply
replace what humans do, but AI can advance in the security industry,
which comes from finding out and unlocking new capabilities that
has never been possible or practical before. This is the true opportunity.
Step back and rethink some of our basic assumptions. What
have we always wanted to do that never felt possible or practical?
Here is one example of the security solutions we’ve been able to
build with AI. In video security where it can be taxing to watch multiple
screens with so much data and activity to view and process, AI can
assist the security team to become more effective. The system is able to
recognize and alert security of suspicious activities like unauthorized
entry, physical violence, loitering and wall-scaling in real time based on
the analysis of behavior-related events captured from the camera. It’s
all done by analyzing spatial-temporal (space and time) information,
such as tracking movement within a video sequence.
Up until recently, we were only able to detect shapes and geometries
drawn on a screen, which did not give enough information to
take action, but we can now take swift action based on knowing specific
behaviors taking place within our actual surroundings. As AI
becomes more capable of understanding what people are actually
doing or intending to do, managers overseeing security are suddenly
able to do more than they ever thought possible in responding to
those behaviors.
Imagine the increased productivity with fewer false alarms and
better response when a security guard actually knows that the alarm
was tripped by someone actually climbing the perimeter wall versus
a person strolling along hitting a 2-D line or a cat darting across the
wall. Imagine the security system capable of recognizing any physical
acts of aggression happening on the floor or large space, alerting
you, rather than having to employ a manual approach like watching
multiple screens and hoping to catch suspicious activity as it happens.
Imagine an alarm monitoring center that does not need to shut off the
alert feeds in bad weather because the AI is able to ignore its effects.
As AI evolves, numerous technical challenges might arise. But the
greatest challenge is actually not technical at all. Bugs can be ironed out.
Systems can be updated. The biggest challenge for the video security
industry today is in expanding the boundaries to attain
what is now possible.
This article originally appeared in the March 2018 issue of Security Today.