16-11-2012, 01:46 PM
A Smart Camera Mote Architecture for Distributed Intelligent Surveillance
Smart Camera Mote.pdf (Size: 2.23 MB / Downloads: 39)
ABSTRACT
Surveillance is one of the promising applications to which
smart camera motes forming a vision-enabled network can
add increasing levels of intelligence. We see a high degree
of in-node processing in combination with distributed
reasoning algorithms as the key enablers for such intelligent
surveillance systems. To put these systems into practice still
requires a considerable amount of research ranging from
mote architectures, pixel-processing algorithms, up to
distributed reasoning engines. This paper introduces an
energy-efficient smart camera mote architecture that has
been designed with intelligent surveillance as the target
application in mind. Special attention is given to its unique
vision system: a low-resolution stereo-vision system
continuously determines position, range, and size of moving
objects entering its field of view. This information triggers
a color camera module to acquire a high-resolution image
sub-array containing the object, which can be efficiently
processed. The paper also presents a basic power model
that estimates lifetime of our smart camera mote in batterypowered
operation for intelligent surveillance event
processing.
INTRODUCTION
Distributed smart cameras have received increased focus in
the research community over the past several months. The
notion of cameras combined with embedded computation
power and interconnected through radio links opens up a
new realm of intelligent vision-enabled applications. Realtime
image processing and distributed reasoning made
possible by distributed smart cameras can not only enhance
existing applications but also instigate new applications.
Potential application areas range from home monitoring,
elderly care, and smart environments to security and
surveillance in public or corporate buildings. Critical issues
influencing the success of smart camera deployments for
such applications include reliable and robust operation with
as little maintenance as possible.
SURVEILLANCE APPLICATION
Our design of a smart camera mote has been pursued with a
specific application in mind: distributed intelligent
surveillance. This guides our design decisions and helps us
in specifying critical mote functionality. We believe that
surveillance will be one of the first areas to benefit from
emerging wireless sensor networking technology.
Especially low per-node cost, ease of deployment,
scalability, and in-network distributed processing are
factors that make this technology ideal for intelligent
surveillance.
Intelligent surveillance may have different meaning to
different people. Let us first consider how surveillance is
typically realized today. Pan-tilt-zoom cameras are
distributed across the deployment area and their raw video
output is streamed to a surveillance center, in which a panel
of monitors displays the video streams. Obviously, this
implementation requires sufficient bandwidth for video
streaming, has high installation cost, and most of all is
hardly scalable. We consider any surveillance solution that
performs processing of the video stream right at the camera
and hence reduces bandwidth requirements as an intelligent
system.
MOTE ARCHITECTURE
The block-level architecture of our smart camera mote
called MeshEyeTM mote is shown in Figure 2. An Atmel
AT91SAM7S family microcontroller [8] forms the core of
our mote architecture. It features a USB 2.0 full-speed port
and a serial interface for wired connection. The mote can
host up to eight kilopixel imagers and one VGA camera
module, for which we chose Agilent Technologies’ ADNS-
3060 high-performance optical mouse sensor [9] (30×30
pixel, 6-bit grayscale) and Agilent Technologies’ ADCM-
2700 landscape VGA resolution CMOS camera module
[10] (480×640 pixel programmable, grayscale or 24-bit
color), respectively. An MMC/SD flash memory card
provides sufficient and scalable non-volatile memory for
temporary frame buffering or even image archival. Wireless
connection to other motes in the network can be established
through a Chipcon CC2420 2.4 GHz IEEE
802.15.4/ZigBee-ready RF transceiver [11]. The mote can
either be powered by a stationary power supply if available
or battery-operated for mobile applications or easy
deployment.
Vision System
The vision system forms the key sensing element of our
smart camera mote. In a first implementation, we will use
two kilopixel imagers with the VGA camera module
centered in between them. All three pixel arrays are parallel
and thus facing the same direction. All three image sensors
are focused to infinity and their field of view (FoV) angle
should be approximately the same although ideally the
kilopixel imagers should have a slightly larger FoV angle.
Hence the three imagers have an overlapping FoV only
offset by the small distance in between them.
We envision the following usage model for the three image
sensors during intelligent surveillance operation. One of the
kilopixel imagers will be used to continuously poll for
moving objects entering its FoV. Once one or possibly
more objects have been detected, position and size within a
kilopixel image can be determined for each object. Basic
stereo vision of the two kilopixel imagers yields the
distance to the object. This information allows us to
calculate the region of interest (RoI) containing the object
within the VGA camera’s image plane. Subsequently the
microcontroller triggers the VGA camera module to capture
a high-resolution grayscale or color RoI including the
detected object. Figure 4 illustrates this hybrid-resolution
vision system overlaid with real images acquired by our
mote prototype. After possibly additional low-level
processing, the object’s RoI will then be handed over to
intermediate-level processing functions.
POWER MODEL
An important performance metric for battery-operated
motes is their lifetime during deployment. The mote can be
powered by batteries deliberately in mobile applications for
instance. Batteries may also serve as backup energy sources
in case the main power supply fails for example due to
intruder attacks. Therefore we developed a power model for
our smart camera mote architecture during basic
surveillance operation.
The power model assumes that the mote is powered by two
non-rechargeable AA batteries (capacity 2850 mAh) at a
conversion efficiency of 95%. It accounts for current
consumption of the following main mote components:
Atmel AT91SAM7S64 microcontroller, SanDisk SDMB-32
MMC card, Agilent ADNS-3060 kilopixel imager, Agilent
ADCM-2700 VGA camera module, Chipcon CC2420 IEEE
802.15.4 transceiver. We used the current draw value for
active and sleep mode quoted in each component’s data
sheet. In the future, we will replace these values with actual
current measurements taken on a mote board.
CONCLUSION
In this paper, we introduced a low-power ARM7-based
smart camera mote architecture. We designed it for realtime
object detection and in-node processing for
applications in distributed intelligent surveillance. Its
hybrid-resolution vision system deploys two kilopixel
imagers to trigger region of interest acquisition through a
high-resolution camera module. We presented a basic
power model, which estimates battery-powered mote
lifetime under varying operating conditions in a
surveillance application. Future work will be directed
towards development of mote boards and research of lowand
intermediate-level processing algorithms.