Energy-efficient operation of GSM-
connected infrared rodent sensor
Gábor Paller, paller.gabor@sze.hu
Gábor Élő, elo@sze.hu
Széchenyi University, Győr
2016 February 19

Precision agriculture and the
agrodat.hu project

• Precision
agricultural
management
• Better control of
production risks and
costs
• Information systems
supporting
decisions
3
Crop production

AgroDat.hu project
Main objective: Establish an
agricultural knowledge centre
and decision support system
• based on data gathered by an
innovative, complex sensor
system and from international
open repositories
• relying on big data, cloud, and
HPC technologies
to support precision agriculture.
Duration: 2014-2017
Budget: 2.4 Mrd HUF (appr. 8
MEUR)
URL: www.agrodat.hu
Consortium:
Local
Senso
r
Globa
l

• Information about environmental and
vegetation processes (supplied off-line)
• Sensors and data capture tools
• Data processing system
• Visualization
• Decision support
5
Elements of the agricultural
decision support system

Sensor lineup
l
2014 – soil and environmental sensors
with simple scalar output – done
l
2015 – camera sensors in multiple
wavelengths with simple and intelligent
triggers – research phase completed,
field-ready implementation in progress
l
2016 – making the camera sensor
mobile

Soil sensor

8
Proposed system architecture
Sensor unit
Sensor
control
Network
Communication
+
App. logic
(Telit GL865)
Data server
GPRS/HTTP POST
requests
WSGI
HTTP server
Database
adapter
Data
visualization
SMSC
End user
Mobile
network
SMS
HP DSM
(Dynamic SIM Management)
SMPP
4.3+
Sensor management
server
Management
user

Server farm

Camera sensor
development

Camera sensors
l
Mission: observe plants and pests in
wide frequency spectrum (visible and
invisible)
l
Simple case: take a series of pictures
and videos at given time spots and
upload them to the server
l
More complicated: detect an event of
interest and upload only if something
interesting is happening

Camera sensor development
l
Use case: common
vole detection
l
Overpopulation due
to mild winters
l
2014 damage
estimation: 500000
tons only in winter
wheat

l
Night animal: cameras in different
wavelength are needed (short-
wavelength infrared and long-
wavelength infrared are being tried)
l
Energy consumption: send images
only if there's high chance of having
animals in the picture → image
processing on the sensor
Advantages of the use case

Common vole in short-
wavelenght infrared
Animal

Common vole in long-wavelenght
infrared
Dynamic mapping to 256-level
greyscale image

Image processing in the sensor
l
Image thresholding – 80% static threshold
l
Get rid of spurious patches – contour
tracing+convex hull filling
l
Close gaps - dilating by a kernel of 6x6
l
Calculate circles of interesting objects –
contour tracing+enclosing circles
l
Find out if the circles move – match the
circles with the circles on the previous
image and flag the circles that don't match

Image processing in the sensor
Eq – input image after
greyscaling
Th – after
thresholding
C1 – after contour
tracing + filling convex
hull
C2 – after dilating
Circle – enclosing
circle

Cost of sending an image
BeagleBone Black (TI Sitara AM335x)+Telit
GL865 modem, image size: 4Kbytes
Architectures:
- GL865 as a modem, all the logic (image
processing+communication) is on the Sitara
CPU – 10.755 mAh
- Image processing on the Sitara CPU,
communication logic on the GL865 – 3 mAh

Power consumption of
communication state machines
l Idle consumption of the Sitara CPU is a
problem in case of action-and-wait type
algorithms (typically communication state
machines)
l If the “wait” phase is long then the
processor could be put into a low-power
mode.
l A separate low-power CPU is justified just for
communication tasks.

Sensor
control
Network
Communication
+
App. logic
(Telit GL865)
Generic model
Same architecture for scalar and
image sensors
Soil sensor
Data acquisition:
ATxmega128u4
Network
Communication
+
App. logic
(Telit GL865)
App. Protocol #1
Camera sensor
Acquisition and
processing: TI
Sitara AM335x
Network
Communication
+
App. logic
(Telit GL865)
App. Protocol #2

Use case categorization
Use case Power consumption balance
between the processing and the
communication activities
Images taken at predetermined
moments of time and sent
No image processing, no balance
Images are taken continuously,
sent if relevant feature is found
Continuous image processing, no
balance*
Images taken at predetermined
moments and sent only if relevant
feature is found
Justified if the image processing
cycle consumes less than sending
the image
* Balance would mean that we compare with continuous image streaming which
is not possible over GPRS due to its low bandwidth.

Cost of image processing vs.
sending
Image processing
(acquiring and processing 5
images)
Sending the image
0.62 mAh 3 mAh
However:
l
Sitara consumes 156 mA in S3 state (69 secs to reach 3 mAh)
l
It takes 4.78 mAh to perform a shutdown-reboot cycle
l
No power saving possible in use case #3 due to high idle
consumption!

Takeaways
l
Low-power, low-consumption devices (e.g.
microcontrollers) are justified even if there's
a full-featured CPU in the system. Typical
use cases: sensor control, communication.
l
No magic bullet: high software engineering
productivity on embedded Linux but high
power consumption too.
l
Look for idle consumption!
l
Linux system optimized for extra-fast
shutdown-bootup cycle may work (but you
need a low-power sensor controller)

Questions?
paller.gabor@sze.hu
elo@sze.hu
24