Artificial intelligence is the talk of the town in most industries
- By Aaron Saks
- Oct 01, 2021
Every technology industry is talking about the
benefits of Artificial Intelligence. More than a
buzzword, AI is hyped as a panacea, while at the
same time, it is often misunderstood by those who
might benefit from it the most.
AI may mean different things to different people, there are
plenty of aspects that apply to all disciplines. The ability for a
machine to “learn” from data it is presented is at the core of all
AI use cases. The term “machine learning” derives from that most
basic idea. Deep Learning, a subset of machine learning, and
is based on neural networks, and is frequently used to analyze
and compare image data. The challenge is how to use these AI
disciplines effectively to better protect people and assets. Beyond
security, it’s time to look at how data from these cameras can be
used to positively impact operations and sales for an organization.
CHANGING HOW WE THINK ABOUT SECURITY CAMERAS
Traditional digital cameras do not identify objects they capture. They
just blindly record pixels to a disk. With analytics, if the camera
sensor detects movement in those pixels, it can place a bookmark in
the recording or send an alert. Anyone who has tried to use traditional
motion analytics, although they can be useful, will know they are
also very prone to false positives. Depending on the installation, a
motion event is triggered by something as mundane as a shadow from
a passing cloud. For this reason, many security professionals shied
away from using analytics in all but the most controlled circumstances
or as a guide to where an event “might” have happened.
Using deep learning algorithms, we can effectively teach a
camera sensor to identify objects and detect unique characteristics
about them. It is a sophisticated process to train a machine learning
algorithm and it can require hundreds of thousands of images to make
it accurate. The algorithms must be told when it gets things wrong, as
well. It is also important to remember that what differentiates today’s
technology from true AI is that machine learning and deep learning
algorithms cannot learn new things by themselves.
Current AI-based cameras can reliably identify objects such as
a car, truck, bicycle, license plate or a person in an image. They can
also discern the unique attributes of these objects, such as color
or whether a license plate or face is present. Thanks to advances
in deep learning, these devices have evolved from capturing
images to becoming highly accurate data gathering tools. They
are network connected, and are truly part of the broader world
of IoT devices that surround us. With their myriad new potential
to protect and inform, it’s time to think differently with regards to
the value these devices can bring to an organization.
SECURITY BENEFITS OF AI CAMERAS
The most common application for AI-powered cameras today is to
empower the traditional motion analytics that we are familiar with,
such as loitering, intruder detection or entering/exiting an area. AI
becomes a powerhouse when used to eliminate false positives from
shadows, foliage or animals, by only triggering the analytics when
the correct type of object is detected, such as a person or vehicle.
The bar is raised further when a deeply integrated AI solution
allows additional descriptive metadata search parameters to speed
forensic investigations, such as searching for clothing color, or if
the subject had a bag, glasses or hat.
AI presents a perfect solution to compensate for unmanned
environments or those with limited staffing, or the loss of
vigilance after looking at a screen too long. AI can help us not
only watch continuously, but also feed systems that are able to
sort, organize and categorize massive amounts of data in a way
that human operators cannot. It can do so far more reliably than
traditional video analytics ever did.
When it comes to protecting assets and people, real-time alerts
generated by a video management system (VMS) enable security
teams to be more proactive rather than reactive as events unfold
in real time. Because AI-powered analytics eliminate false alarms,
they can more accurately determine incidents that require further
investigation by operators. Thanks to the extra data AI-based
cameras can capture, analytic rules can be enhanced with more
sophisticated logic and customization for precisely what an end user
requires. For example, we can tell a camera to ignore all cars, but
to alert us when a person comes to the door. AI can help us count
objects like people or cars more precisely than ever. This includes the
ability to count objects accurately even when they partly “occlude”
or pass in front of each other. This is key since it allows use cases
like people counting from more sensible camera view angles. This
is far superior to conventional people counting techniques, which
require a top-down view to avoid occlusion, and which give a less
useful camera view when you want to identify faces as well.
POST-EVENT FORENSIC SEARCH
When it comes to post-event forensic searches, AI-based cameras
are in a league of their own. Additional descriptive metadata
about objects is captured within each frame. Because the
metadata is small, it adds very little to the overall bandwidth and
That metadata, which might include descriptive characteristics
of objects like the color of a person’s shirt or pants or their
approximate age and gender, enables a VMS operator to quickly
search through video to find a particular object or person. A search
that might have taken security staff hours or days to complete now
takes only seconds when the search includes additional metadata
provided by an AI camera.
GOING BEYOND SECURITY
Although people counting, heat maps and queue management
analytics have existed for some time, they too have been subject
to the inherent inaccuracies of pixel-based motion detection.
Conversely, AI-based object detection delivers profoundly accurate
data and metrics for operations, sales and marketing teams looking
for insight on everything from retail store performance to ensuring
process efficiency and operational compliance.
As a result, these cameras have become an indispensable tool
for business operations. Depending on the business, the value
proposition for such data can be a game-changer worth many
times the cost of the system.
For customers with more sophisticated data analysis needs,
camera metadata is accessed and combined with other data. It
is processed by other platforms for sophisticated visualization
and data mining. This allows technology partners to access the
aggregated data into their own charts, graphs and exception
reports powered by specialized software they may already be
using. There are familiar use cases spanning multiple industries
that require linking data from access control, intrusion, point of
sale systems, staffing data, schedule data, weather data and many
other data sources. The potential for this unified data to create
comprehensive business solutions is substantial.
Since video cameras are already well accepted and commonplace,
the opportunities for them to evolve into unobtrusive, important
data gathering tools for business and operations intelligence will
only continue to grow. Operations and marketing departments
may also find common ground when budgeting for a system that
can serve the needs of both departments.
BENEFITS OF AI ON THE EDGE
AI-based analytics can run on the edge or in a server, but there
are significant aspects to each method of deployment that should
be considered. With AI on the edge, valuable events and other
metadata generated at the camera must be gathered from many
endpoints and that data must be aggregated together to enable
clear visualization of the trends and anomalies identified. This
can be done on a lightweight local server that also runs the VMS.
Running AI analytics at the camera significantly reduces the
overall cost of the equivalent server resources required to run
AI-based analytics since edge-based analytics run before video is
compressed and streamed.
Running AI on a server requires that the video stream be
first decoded which requires CPU/GPU resources that can scale
dramatically as stream count increases. While the power of a
server far outweighs what a camera can provide, there is a point
of diminishing returns when electing to do everything on a server
for all, but the most demanding processing. For that reason, a
hybrid approach, in which AI analytics are performed on the
edge and the lightweight data results are sent
to an inexpensive server or workstation for
aggregation and display, will remain a popular
choice for some time.
This article originally appeared in the September / October 2021 issue of Security Today.
Aaron Saks is the product and technical manager at Hanwha Techwin America.