Oliver Vellacott, CEO of IndigoVision, warns against setting too high an
expectation for what video analytics can deliver.
Analytics will detect 'suspicious movement' from people walking along a
street. Analytics will detect terrorists walking round a hillside a mile
away. Analytics will pick an offender out from a sea of faces. These are
just some of the misconceptions about analytics today. Was there ever a
technology that was so "over-promised and under-delivered"?
No, the reality is that analytics are still very much in their infancy.
Expectation management is at the heart of the issue being realistic with end
users about what can be achieved. The fundamental problem is that as humans we
do these tasks without even thinking about it. We read license plates and
recognize faces totally subconsciously. It may have taken us years of learning
during childhood to acquire these skills, but the fact is that we take them for
granted.
Computers, on the other hand, lack even the basics of visual intelligence.
They can perform some video analytic functions reliably, but often only by
severely constraining the application. Qualification is everything, and setting
the end user's expectations is absolutely vital.
Analytics Today
So, what can be done today? License plate recognition has been around a long
time and is well proven, however it is still not 100% accurate. Face recognition
is notoriously difficult to perform reliably, and is extremely easy to fool by
using disguises and to work with any degree of accuracy an excellent headshot of
the subject is required.
There are however some bread-and-butter analytics functions which CAN be
performed reasonably well today. Motion detection is the simplest, most basic
form of analytics. Many manufacturers support it but very few systems achieve
sufficiently low false alarm rates to be useable. A system which generates
anything in excess of 20 false alarms in a night becomes ineffective, because
all alarms very quickly become ignored or the motion detection gets switched off
just to prevent the operator getting "drowned in alarms"! Aesop unwittingly
foresaw the reality of most security systems when he wrote "The boy who cried
Wolf"!
Congestion detection has evolved from basic motion detection. When the
density of humans or cars reaches a certain level an alarm is triggered. Counter
flow looks for objects which move 'against the flow', and is valued in
applications like airport security. Virtual tripwire is also a refinement of
motion detection in that it triggers an alarm when someone or something 'breaks'
a line that has been drawn in the image. This is useful in applications such as
large areas with 'no go zones', such as in factories. People are allowed to
happily move in 'free areas' but the system alarms as soon as any of them move
out of these see sidebar analytics A to Z(at end of this document).
With all these analytics applications, camera positioning, lens selection and
lighting are absolutely critical. Just changing the camera position can easily
improve the analytics performance by an order of magnitude. For example, in
counter flow detection the algorithm has a far easier job if the camera is
positioned pointing down so that it sees the area in 'plan view'. If it is
looking from a perspective view - with human bodies occluding one another, the
process of tracking people becomes much harder, and expectations can once again
become unrealistic.
Crowded Marketplace
There are literally hundreds of companies touting analytics software. As a
manufacturer of complete IP Video solutions, IndigoVision gets approached by on
average one analytics provider every week, asking us to integrate their product
into our IP Video management platform. Why so many? Because any small software
company can develop a suite of video analytics by buying a framegrabber, a
powerful PC and writing some software.
These are then sold as separate, stand-alone systems which sit 'next to' the
main CCTV system. Video is split from the matrix and fed into the analytics
system. This has the limitation that it is not truly integrated into the
operation of the main CCTV system and therefore delivers only limited benefits.
Analog just doesn't really support integrated analytics. DVRs do to a greater
extent, but they still remain largely 'islands', i.e. not integrated, just as
with analog CCTV.
Integrated Analytics the IP Video Solution
IP-based video management systems provide the ideal platform for powerful
analytics to be completely integrated into the system, making them a core and
integral part of its operation. Leading IP Video solutions support analytics
that can be performed in two fundamental modes: live (to detect events as they
occur) and post processing (to test various scenarios on recorded footage).
The optimum place to locate live analytics is obviously at the camera, as it
is the only truly scalable solution and also doesn't use up network bandwidth.
Central real time processing will eventually run out of steam, whereas every
camera can have dedicated processing. For example a camera with built-in
analytics can monitor scene activity and transmit only on specified events (e.g.
a person moving the wrong way through airport security). This reduces
unnecessary video traffic on the network, thus reducing the bandwidth
requirements and simply cannot be done with traditional analog.
The optimum place to locate post-processing analytics is obviously on a
central server, so that recorded video can be searched many times, with
different parameters. One of the biggest time wasters for operators used to be
fast forwarding and rewinding through VCR tape. It improved with DVRs, but most
are still basically a digital fast forward or rewind. Analytics offers the
potential to further transform this essential task by searching large amounts of
recorded video for possible events and having the operator validate them.
Computers do what they are good at - locating possible events - and humans do
what they are good at - verifying those events.
Summary
Setting expectations is everything. It's important not to believe all the
ridiculous hype and nonsense about what analytics can deliver. In 30 years time
it will probably be possible, but today it's all about making sure that what is
achievable is done extremely well.
| ¡°Analytics
A to Z¡± |
Abandoned Object
Detection
¡¡ |
Live: Used
for alarm generation when an object has been left in a busy scene
(such as a suitcase in an airport or railway station), this feature
is a key component in the timely management of dangerous situations.
This functionality can also be used to detect illegal parking or
vehicles staying too long in certain zones.
Recorded: It can also
be used to search recordings for events such as parking violations
and blocked freeways. |
Congestion Detection
¡¡ |
Live:
Congestion Detection is used to alert a user in the event of a
build-up of congestion in an area of interest (railway station
platforms, public spaces, motorway entry/exit slip roads,
point-of-sale queues, etc.). This helps to initiate timely action
and prevent an undesirable situation from worsening.
Recorded: It
can also be used to provide statistics for staff planning and
marketing purposes. For example, it can detect when a shopping mall
is at its busiest, or when hypermarket queues start to build up. |
Counter Flow
¡¡ |
Live:
Counter flow is available to alert a user to a person or vehicle
moving in an unauthorized direction, such as a person moving against
the permitted flow in an airport immigration or customs area, or a
vehicle traveling in the wrong direction on a carriageway or in a
one-way system. Recorded:
Counter flow analysis can help optimize crowd control in public
areas, such as the underground, or train stations. |
Motion Detection
¡¡ |
Live:
Motion detection can be used to alert users of unauthorized entry,
and of potentially dangerous situations, for example, if a member of
staff is entering a hazardous area without protective clothing.
Recorded:
Users can define specific areas of interest in a scene and search
automatically through a recording to identify and view any
significant motion that occurred during the recording. This is
hugely useful when searching for events in corridors, staircases,
walkways, etc. during quiet times. This can be tuned using
parameters such as object size and sensitivity. |
Shape-based
Detection/Object Tracking
¡¡ |
Live:
Shape-based detection/object tracking can be used in a wide variety
of applications. It can alert CCTV operators when a high-sided
vehicle or ship approaches a low bridge. Alternatively, it can be
used to distinguish between an animal approaching a boundary fence,
and an intruder. Recorded:
In recorded video footage, this feature can be used to analyze the
types of vehicles using a road, and what time of day they are using
it. |
Theft Detection
¡¡ |
Live:
Museum mode can be used to detect theft, such as the removal of a
painting from the wall of an art gallery. In this mode sensitivity
is configurable and moving foreground objects are ignored.
Recorded: It
can also be used when reviewing recording footage, for example of a
warehouse or a stockroom. It can quickly identify when a particular
item was moved or removed from the scene. |
Virtual Tripwire
¡¡ |
Live: With
a virtual tripwire set alongside a railway track, freeway hard
shoulder, building perimeter or around a temporarily parked asset
for example, the operator will be informed when that tripwire is
breached. Since the system ¡°understands¡± direction, alarm
discrimination based on direction of approach is made possible.
Recorded: A
virtual tripwire can also be placed on the entrance into a building
or parking lot to review how many people or vehicles enter it. |
¡¡
HIGH-end Video Server Module board W1860
