Beyond the Basics
Understanding the analytics behind the networked security camera
A security camera is much more than a security camera
when it’s part of an analytics infrastructure that
serves completely new goals, advancing the organization
and management of businesses and even entire
cities. Many end users are surprised at how much data
can be mined from an innocuous camera on the ceiling or on the
street corner.
Just as computers and VMS continue to get “smarter,” so too are
the cameras themselves, evolving their capabilities to see beyond the
pixels they capture. Analytics on the edge is nothing new, and most
respectable security cameras have a variety of ways to detect motion
and basic behaviors of objects in the frame. They can detect loitering,
direction of travel and can specify zones to pay attention to.
The latest cameras also detect sound events and can warn users
of gunshots, screams, explosions, and glass breaks. These are all very
valuable from a security standpoint and will continue to evolve in
the coming years. What many don’t realize is that the ability to completely
customize a camera to become a business or city management
analysis tool is also making great strides.
Giving Cameras a New Job
Basic edge analytics for security are typically built-in to today’s security
cameras. Going beyond security, there are a number of specialty
analytics such as license plate recognition (LPR), traffic, people
counting, and heat maps that can be made available to provide new
levels of business intelligence and city management. Buying a camera
that is purpose-built for business analytics or LPR is possible, but not
always practical.
What if you want your LPR camera to also count people? The
answer is to let end users choose the analytics they require from a
collection of best-of-breed third-party developers. This has the benefit
of reducing the material cost of manufacturing specialty cameras
and also simplifies the installation and future upgrade potential. Basically,
we want to design a camera for all seasons where users can
install custom analytics applications as simply as one installs an app
on a smartphone.
These applications have access to the raw video coming from the
sensor and can analyze it onboard of the camera. At Hanwha, we
call this capability the Hanwha Open Platform and it is a core feature
of our Wisenet X and P series cameras. Hanwha engineers have
purposefully reserved space in our Wisenet 5 chipset to allow open
platform applications from third parties to be installed as required by
the end user. Not only does it unlock powerful new tools and features,
it also enables customers to pay for only the analytics they need, when
they need them.
This capability, while not exactly new, is seeing rapid adoption as more integrators and end users discover the business value inherent
in what was previously a security expense. A security system that
delivers actionable intelligence about businesses and city infrastructure
can rapidly pay for itself and even become a revenue generating
tool in some instances.
Best-of-Breed
So, what types of analytics can third-party vendors provide today?
Most people are familiar with license plate recognition cameras and
servers. Depending on the requirement, an application like Arteco
LPR can integrate seamlessly into a Wisenet X series camera providing
license plate recognition at speeds up to 80 miles per hour. It can
work very well for traffic management, parking management, access
control, vehicle tracking, and crime prevention.
For a different type of traffic monitoring, the Automatic Incident
Detection (AID) application can detect stopped vehicles, slowdowns
and congestion, wrong way drivers, pedestrians, smoke/low visibility,
and lost cargo. The application collects data on traffic flow such as vehicle
counting and classification, traffic density, average speed, and more.
In the retail space, business analytic applications like those from
RetailFlux enable people counting, heat maps, shopper flow, route
maps, zone analysis and queue monitoring. No records are kept
about shopper identities, only actionable insights about what is working
and what needs improvement.
Most of these analytics applications also come with their own
front-end software or server to collect and display the data in a way
that is useful for end users and stakeholders. Integration with popular
video management systems is common as a way to dashboard the
information display for ease of use.
The Future Looks Bright
for Analytics and Metadata
All of these applications share one thing in common: They collect
metadata. Metadata, at its core, is simply data about data. As technology
continues to evolve, metadata collection and analysis will become
a more important and necessary part of the successful management
of cities, businesses, and security.
Combing through surveillance video for an event that occurred
sometime in the last 48 hours is a laborious task for anyone. Today,
metadata generated by the camera, whether it be motion or sound
trigger marks, helps staff find events quickly.
In the future, as we see new developments in AI and Deep Learning,
we can expect these on-edge analytics to take a big leap forward
in the types of metadata they are able to extract. Instead of simply
marking up motion events, future applications will utilize object
classification techniques to detect physical objects like a chair, car,
bus, motorcycle, train, human, cat or dog. If you have a little bit of
information about the scene—red car, person wearing black pants
and green jacket, dog barking—this will enable VMS to quickly find
footage of interest, all thanks to the rich metadata being captured by
the cameras.
Another aspect of deep learning is about how we store and restore
compressed video data. Storing RAW footage from the camera, particularly
as greater than HD formats become common, is not practical
due to immense file sizes. We already compress/encode into H.264
and H.265 today because of need for saving bandwidth and storage
space. By the very nature of compression, we are discarding some
data in the image that is typically not important to how the eye sees.
With deep learning, we can imagine how an AI algorithm can learn
how we have encoded and compressed an image and then be able to
restore even compressed areas of an image during live viewing and
critical analysis.
The best thing about technology is that it never stands still. The
applications-based model for installing purpose-built on-edge analytics,
on demand, is here to stay and should be
an important consideration as you plan for future
deployments and upgrades to existing security infrastructure.
This article originally appeared in the May 2018 issue of Security Today.