September 2016
38
I
maging
& M
achine
V
ision
of the hardware features in a single, unified
environment. The latest generation of AMD
SoCs represents how the theory of HSA is
put into practice. Heterogeneous process-
ing platforms essentially combine processors
with different architectures in a single device,
which AMD refers to as the Accelerated Pro-
cessing Unit (APU). This brings together the
powerful x86 CPU and its ability to efficiently
execute sequential programming, with its
highly optimized Graphical Processing Unit
(GPU) designed to tackle parallel processing.
By harnessing the potential of the GPU for
intelligent vision systems, the HSA delivers
greater power efficiency and, therefore, a total
increase in processing capability.
The software ecosystem for advanced appli-
cations like intelligent vision systems also
benefits from the HSA approach. Open
standards such as OpenCV (Open Source
Computer Vision) and OpenCL (Open
Computing Language) make software devel-
opment much simpler and so harnessing the
power of an HSA less challenging. Develop-
ers can create advanced software applications
without needing to focus on partitioning
code between the various processing ele-
ments in an HSA; the platform and low level
software has been developed to help with
code partitioning and execution, in order
to get the highest performance. This allows
many powerful and complex algorithms to
be efficiently ported to an HSA.
As intelligent vision systems develop and are
deployed in significant numbers to enable the
next generation of robotics, society will rap-
idly come to rely on them. Perhaps the most
apparent of these applications will be auton-
omous vehicles, but their use will be wide-
spread. Their safe operation will be imperative
and the integrity of the data stored essential,
because decision making (a fundamental fea-
ture of advanced robotics) will rely heavily on
the integrity of the data available. There would
be no human operator present to blame, for
example, if an accident occurs due to the mal-
function of a care-giving robot.
Highly dense integrated devices are increas-
ingly prone to the phenomena know as single
events. These occur when a single ionizing
particle comes into contact with a transistor
or other integrated element causing a change
of logic state. The result, termed a single event
upset, can change logic 0 to logic 1 in a mem-
ory device, for example, thereby changing the
way a piece of code executes or, in other words,
the decision an intelligent system makes. Pro-
tecting against the effects of single events is
taken extremely seriously by the aerospace
industry due to the potential consequences
and, in part, to the closer proximity of the
electronic systems to the cause: cosmic par-
ticles colliding with atoms in the atmosphere.
While the probability of a single event occur-
ring varies between systems, they could poten-
tially occur once every 100 hours of operation.
Protecting against the effects of single events
hasn’t been a high priority for most applica-
tions outside the aerospace industry, but for
machine vision systems it could become so.
Independent tests carried out by the NASA
Goddard Space Flight Center have demon-
strated that the AMD G-Series SoC can
withstand a total ionizing radiation dose of
17Mrad(Si). This compares favorably with
the specification normally applied for stan-
dard space flights, of just 300krad. While a
single event could hit anywhere, integrated
memory such as SRAM is most susceptible
due to its high density, while its importance
to code execution is apparent. AMD employs
advanced error correction (ECC RAM) in the
G-Series SoC, which is able to compensate
for the effects of single events. This makes the
G-Series even more applicable to intelligent
vision systems.
The development and deployment of robotic
systems is set to extend beyond industrial
automation to all vertical industries. Intelli-
gent vision systems are a key component of
that evolution, forming one aspect of the three
technological trends fuelling the evolution of
robotics; high resolution sensors, powerful
heterogeneous system architectures (HSAs),
and highly efficient brushless DC motors.
HSAs like the AMD G-Series SoC offer a
highly integrated and powerful combination
of processor architectures in a unified plat-
form. When coupled with open source soft-
ware ecosystems, developers are empowered
to create solutions that allow more advanced
deep-learning algorithms to be used, moving
machine vision systems and robotics from a
sense-compare-decide model to a sense-plan-
act behavioral scheme. Machine vision sys-
tems like the intelligent vision system IVS-70
from Unibap, enabled by the AMD G-Series
SoC and Microsemi SmartFusion2 FPGA,
demonstrate how advanced integrated devices
and software are coming together with high
performance optical systems to deliver
machine vision solutions that will empower
the robot revolution.
n
Figure 2. Time synchronization is perfect at the microsecond level, even if the computer screen
can show only milliseconds. Picture taken with a 10 Mpix stereo-pair (2 x 5.2 Mpix) of 70 mm
lenses.
