Images, such as those from closed circuit cameras, are routinely used in security systems, despite being of poor quality and laborious to interpret. Machine vision provides methods to enhance picture quality interpret events and monitor complex scenes.
Cameras monitoring scenes, such as prison perimeters, are often linked
into the alarm system. If the alarm is triggered, images are stored
allowing the cause to be identified. Many false alarms are caused by
wildlife, and much time is wasted following them up. A system to locate
and identify the causes of alarms can be used to minimize the number of
false alarms. The unobstructed view from the camera is used as a
reference image. If an alarm is triggered, the stored images can be
analysed by subtracting them from the reference image (Figs 1 and 2).
The resulting image will locate the intruder (Fig 3). Using information
regarding shape, movements and speed of the intruder, and its position
on the scene plan (Fig 4), the system can identify harmless intruders.
When wildlife, such as rabbits and birds, are positively identified, the
system will prevent the alarm sounding.
Machine vision techniques have made it possible to read the registration
number of a vehicle from a video image. This technology can be used in
many appplications including stolen vehicle recovery and journey time
analysis. This information can be used locally, or sent via
communication channels to a database or control computer. Reading
number-plates is more economic and less invasive than other techniques
of vehicle monitoring, such as using transponders, and has significant
operational advantages. Four stages of processing are involved. Firstly,
the number-plate is detected and located in the image. Next, the plate
area is analysed in detail, and regions that could be characters are
extracted. The extracted characters are each then matched against a set
of templates. The statistical results from this process are used in the
final stage, where registration number formats and syntax rules are used
to derive the output result.
In some situations the movements of people may be of interest, e.g. when
monitoring shopping precincts. In such cases, a model-based system can
be used to locate and track people in video images (Fig 1). The edges of
the objects in the image are found (Fig 2), and then, using a model,
anyone in the picture can be located. The model comprises 16 articulated
cylinders which represent the major body parts. Once a person is
located, they can be tracked by predicting and testing the possible
movements from a range of legal articulations of the model (Figs 3 and
4). The components of the model are proportionately sized with ther
actual sizes depending on the height of the located person. This system
is being developed to monitor a scene and learn the usual paths taken by
individuals. Then any deviation from the usual paths can raise an alarm.