Candid Cameras: Setting up
wireless networks for surveillance and beyond
 |
The CEPANS lab group
works with wireless sensor networks. Back row (from left): Professor
Seapahn
Megerian, Yen Ting Lin, Hsiang-Kuo Tang, Steve Myers. Middle
row: Tai Hsuan Wu an Chao Wang. Front row: Kamal Srinivasan. (Not
pictured: Jake Adriaens.)
(View larger image)
|
urveillance is not a new concept to military
or security forces. In an age of webcams and cell phone video, it would
seem that creating a wireless video surveillance network to monitor
a field would not be difficult.
But there are many challenges that arise when creating
a network in the field. If the deployment is random, it is hard to control
where the cameras are facing or where blind spots occur. Even with controlled
placement, there is the possibility for improper orientation, such as
a camera looking at a wall, or redundancy, such as two or more cameras
covering the same area. With the volume of information constantly uploaded
from the cameras to remote computer stations, orientation and redundancy
create problems. Then there is the problem of power: Since each camera
needs energy to run, having all the cameras on all the time isn’t
practical.
These challenges are the focus of the Collaborative
Embedded Processing, Actuation and Networked Sensing (CEPANS) Lab, directed
by Assistant Professor Seapahn
Megerian. The group’s goal is to discover methods that will
optimize sensor coverage while reducing the equipment and power required.
“We’re looking at mathematically solving the problems of
what’s the best orientation, what’s the best location, and
which cameras should be on to guarantee that I don’t miss something,”
says Megerian.
For example, one problem CEPANS is working on is calibration,
such as correcting for position error. Even with Global Positioning
System to help with sensor placement, there is error.
“I might think the camera is on one side of a wall, when it’s
actually on the other side,” says Megerian, “so in calibration
I’m correcting for this position error that could lead me to draw
incorrect conclusions.”
Calibration is especially important with detection
sensors, or cameras that are activated by motion or objects in their
field of view. When part of a network, cameras in the same area should
report the same input, but if not calibrated properly, a camera might
fail to identify a target in the area or may activate falsely.
Cameras are not the only type of sensor that CEPANS
works with. Sound, light and temperature sensors can also be networked
for surveillance in monitoring. However, they have a different type
of coverage area than cameras do. A microphone may pick up sounds in
all directions, so a map of its coverage would be a circle. But many
cameras are directional, with cone-shaped field of view. “Based
on this coverage model, we formulate questions, such as, how well do
I see a field when I have stationary sensors?” says Megerian.
 |
Grad student Jake
Adriaens sets up a camera network at a DARPA demonstration showcasing
the group's low-power wireless video surveilance networking prototype.
The cameras lined up on the table send video feed to the laptop,
monitoring movement of a robot across the room.
(View larger image) |
The CEPANS group answered the coverage question for
omni-directinal sensors a few years ago, but recently came up with a
new algorithm to detect breach, or unauthorized movement, in camera
sensor networks. “With our algorithm, we can very efficiently
calculate what is the best an ‘adversary’ can do in trying
to traverse this field undetected,” says Megerian. Specifically,
it can calculate how close the intruder has to come to a sensor, even
if staying as far away as possible.
Charting coverage and possible breach routes helps
with many aspects of deploying a sensor network, such as number, placement
and orientation. The algorithm is one step to determine how to best
place the fewest number of cameras for maximum coverage and minimum
redundancy.
There are many obvious military applications of this
technology. If the military wants to monitor a sensitive area, it can
use the algorithm to figure out if it has enough sensors to cover the
area or if there is a route that an enemy could take through the area
and be undetected. They can also monitor areas where eyewitness surveillance
is not possible.
However, wireless sensor technology has many applications
in the civilian realm as well. “Anytime you need to monitor some
area, security or surveillance anywhere, either private or commercial,
there are applications for this. Wildlife habitats, security, traffic,
even space science—there are endless possibilities,” says
Megerian.
“One thing that is new in this area is that
sensors are becoming very tiny, very cheap, and low-power enough that
they last for a long time,” says Megerian. “Video cameras
have been around a long time, but you couldn’t afford to put a
hundred of them in a field. They’re becoming cheap enough that
the commercial applications are growing.”
Those commercial applications include security for
places like casinos or banks or quality assurance surveillance in factories
and warehouses. Wireless sensor systems also can contribute to customized
workplace environments. For example, a system could be set up to give
employees their preferred light and temperature settings at their desks
or workstations, says Megerian. The more possibilities that are opened,
the more sensor network technology could be developed. “We are
driving applications, which creates more need for technology,”
he says.
