Learning to Adapt
Video analytics plays a key role in revolutionizing physical security
- By Eric Eaton
- Dec 01, 2008
A new type of video surveillance technology called adaptive
learning video analytics is revolutionizing physical security
with its ability to analyze input from hundreds or even thousands
of security cameras and provide alerts to potential threats as
they occur. Conventional video surveillance systems are often ineffective,
since they rely on people to monitor all the cameras’ output,
which is a virtual impossibility. Some solutions use rules-based algorithms
to analyze video output and detect one specific behavior—but
human behavior is too various for this approach to be effective. And
neither approach provides alerts in real time.
Adaptive learning video analytics takes a different approach by
analyzing the output of video cameras in real time to detect—and
alert on—abnormal behavior. Because this technology is computerbased,
it brings physical security into the realm of IT like never
before. In fact, IT will play an important role in implementing and
managing this next generation of video surveillance systems and thus
will need to learn about these systems and how they work.
Intelligent Technology
The basic concept of this new technology relates to other, wellunderstood
technologies that IT departments already use, such as
software-based performance management and network security
products that employ pattern recognition. These products analyze
a system’s performance over time to learn normal patterns of
activity and recognize activity patterns that are abnormal, without
human input. They also send alarms that alert users to any
detected abnormal activity.
A system’s performance management product might learn,
for example, that there is normally a spike of e-mail activity at
8 a.m., as people start their day’s work, but that a spike of email
activity at 4 a.m. is not normal. This abnormal activity,
which could represent an attack by a worm or virus, will trigger
an alert.
The Core of Surveillance Technology
Adaptive learning video analytics uses similar principles.
However, it has extra layers of complexity because its roots are
not only in intelligent pattern recognition but also in observations
made by video analytic algorithms. Video surveillance systems
are designed to minimize threats from people, but people’s
behavior is unpredictable. It is impossible to write a set of rules
that can be expected to cover the full range of possible behaviors
for any given environment. Instead, the system must be able to
learn what is normal and what isn’t.
In essence, adaptive learning video analytics follows the
human cognitive model for processing visual inputs into
knowledge and requires a combination of computer vision,
video analytics and machine-learning capabilities. The technology
takes the input from existing video security cameras—
the eyes of the system—recognizes and identifies the objects
in each frame to learn what activity normally takes place within
the area under surveillance; analyzes the changes, activities
and motions of those objects; and builds a model of established
behaviors.
Finally, adaptive learning video analytics provides a wide
range of alert systems that can raise awareness and report abnormal
or high-risk behaviors. This is possible because the system
can compare current behaviors to patterns it has learned through
observation. All these activities take place automatically, without
the need of constant human involvement to create rules and
update settings whenever a camera is deployed to a new location.
Unlike programmed rules-based solutions, adaptive learning
video analytics can continuously improve on its accuracy and
utility by adapting to changes in the observed environment,
thereby also improving the value of the technology. This constant
self-calibration and self-improvement enables the technology
to continuously provide accurate analysis of potentially
threatening behaviors.
Physical, IT Security Strategies
Adaptive learning video analytics accelerates the convergence of
strategies for physical security and IT security. The technology is
computationally intensive and thus has hardware and networking
requirements that fall into the domain of IT.
Computing environment. The new adaptive learning video
analytics technology is run on a series of servers that need to be
deployed in a data center environment, complete with appropriate
power and cooling requirements. These systems also must
maximize CPU and memory utilization, and they are best suited
to running on lean operating systems such as Linux.
Protocols. When evaluating video surveillance systems, it
is important to remember that many employ proprietary protocols
to perform functions such as moving video streams and
are therefore often not compatible with existing IT infrastructure.
Adaptive learning video analytics embraces open
standards, employing the real-time streaming protocol for
communication of video data and other standard protocols
such as lightweight directory access protocol, which makes it
much easier for IT to add the infrastructure required to support
video surveillance.
Use of other Web service protocols such as extensible markup
language and simple object access protocol also means the technology
can integrate with existing IT networks to correlate all
types of security alerts.
Integration with IT. Because video is bandwidth intensive,
IT must be able to provide sufficient bandwidth between the
servers handling adaptive learning video analytics and other
enterprise servers. It also is important to adhere to industry standards
regarding data compatibility and to be up-to-date on how
they are evolving. Video compression standards have improved
greatly in recent years, so it is always best to insist upon modern
video compression encodings such as MPEG-4, H.263 and
H.264. These protocols will save significant bandwidth when
building a large deployment of cameras integrated with adaptive
learning video analytics.
Architecture. An effective video analytics system must be able
to communicate and interoperate with business logic and data, so
it is important to employ a Web services platform architecture.
As more physical security solutions continue to incorporate
intelligent software to augment the capabilities of their operations,
it will become natural for these solutions to migrate under
the domain of IT. In addition, as physical security continues to
evolve from “guns, gates and guards” to using a wide variety of
intelligent technologies, it also will eventually be migrated into
and managed as an additional component of an organization’s
overall corporate IT security model.
The sooner IT begins learning about
how these technologies integrate into its
domain, the sooner these technologies can
be leveraged to benefit the enterprise as
a whole.