Perhaps
the newest, most commonly discussed and advertised technology in
machine vision in recent times is Embedded Vision.
But what is it actually? And why has it become so prevalent suddenly?
The easiest way to answer this is to go back in time.
From the beginning
Historically, machine vision systems have consisted primarily of
a camera connected to a computer via some form of analog or digital
interface.
Initially, the machine vision industry used CCTV camera technology
for image sensing. Computers were fitted with analog frame grabbers
that supported cameras with PAL, NTSC and any of the other CCTV
formats. The image was captured by the frame grabber, transferred
to the PC’s memory and there, the image was processed by software.
The software might perform a number of different taks: enhance the
image, analyse and identify features in the image, make measurements
on those features to produce an inspection result.
Then came the progressive scan digital cameras that used the LVDS
interface, then Firewire cameras that harnessed the IEEE1394 interface
that was being built into some computer motherboards. Later USB2
was another ubiquitous PC interface that was leveraged by the machine
vision industry to connect to cameras. Then along the way, as resolutions
and speeds increased, we saw others emerging; Cameralink, Cameralink
HS, Gigabit Ethernet, Coaxpress and the different revisions of these
as they developed further.
Digital machine vision cameras over the years have become progressively
more sophisticated with in-built functionality to process and improve
the quality of the image. This removed the burden from the PC software
and enabled the complete system to run faster. Features like defect
pixel correction, flat field correction, colour conversion, gamma,
multiple regions-of-interest and many others added value to the
camera but inherently also increased its cost.
Regardless of the camera technology or the interface used, the humble
PC has been the lynchpin of machine vision systems for a long time.
Intel processors have featured heavily and with all its warts and
short-comings, Windows has been predominantly used as the operating
system.
In more recent years, we have also seen the advent of smart
cameras where the image processing, analysis and measurement
is done within the camera. Several companies like Teledyne
Dalsa , LMI and others
offer smart cameras with CPU’s, FPGA’s and GPU’s
to accelerate processing. They typically have an in-built user interface
for the operator to interact with the camera to setup the inspection
and measurement task. Although these cameras are quite powerful
and configurable, they are not flexible enough to perform an extremely
wide set of tasks, and they come at a cost.
Where embedded vision is today
So here we are today still relying on the PC, Windows or on smart
cameras - and this is where embedded vision steps in.
Due to the relatively recent emergence of very powerful, low-cost,
and energy-efficient processors, it has become possible to incorporate
practical computer vision capabilities into embedded systems, mobile
devices and the cloud. It’s expected that over the next few
years, there will be a rapid proliferation of embedded vision technology
into many kinds of systems.
Where embedded vision will be in the
future
But other than the low cost of these powerful new processors, what
else is driving the market to take-up embedded vision technology?
The influence of IoT and Industry 4.0
I believe that IoT and Industry 4.0 are contributing factors as
the ability for embedded vision to distribute the processing to
the image source fits well with IoT and Industry 4.0. Rather than
having multiple image sensors pushing high bandwidth data back into
a central powerful PC, embedded vision allows for image processing,
analysis and measurement at the sensor and so outputs lower bandwidth
data that can then be transferred to the cloud.
Autonomous vehicle technology
Technological advances and the transformation from driven to autonomous
vehicle technology is another driver of embedded vision. Autonomous
vehicles have a requirement for a multitude of image sensors to
provide the necessary feedback for guidance and safety and there
is a great deal of development currently happening in this area.
The machine vision industry has been very good at piggy-backing
on advances in technology made in larger industries and this is
another example.
Portability and Flexibility
Another driver of embedded vision, is the ability of manufacturers
to build cameras with different levels of sophistication, offering
greater flexibility and cost savings.
An example is Allied Vision’s Alvium camera series, one of
which is presented in the image above. This camera range offers
the very simple low-cost board-level Alvium 1500 cameras that do
only very basic on-board processing and then connect to a low-cost
embedded processor card to do the rest. They use MIPI-CSI2 to connect
the camera to the embedded processor with a standard video interface
for embedded vision. Where the application requires more processing
on the camera, the Alvium 1800 or 2000 series offer a mid-level
solution where some of the image processing is done on the camera
and the rest done on the embedded processor card. The Alvium 3000
series will do advanced on-camera processing. The Alvium 3000 cameras
are much lke the fully functioned cameras we have today with USB3
and GigE outputs.
This modularity offers a good platform to move across
as a developer’s needs change as it allows the developer to
design-in a camera that best fits cost and functionality requirements.
Which Embedded Processors
So what are these low-cost, embedded processors we talk about?
To be honest, low-cost is not always the case. There are some embedded
processor cards that cost as much as a PC but they typically offer
very significant performance benefits – for example up to
10x the graphics capability of a PC.
Some of the more common embedded processors being used are the i.MX
family from NXP which feature the ARM Cortex CPUs and run Linux.
The Jetson TX series is another ARM Linux solution along with the
Jetson Xavier with ARM and lots of nVidia GPUs.
The Embedded vision market is I believe still in a state of flux
as camera manufacturers try to establish their technology as the
mainstream. I think the takeaways right now are that machine vision
is heading towards a more distributed processing form, ARM and Linux
will gather momentum and cost will be at the forefront of most selection
decisions.
Marc Fimeri
Managing Director
Adept Turnkey Pty Ltd
|