About six months ago, a camera manufacturer inquired whether Alacron
(Nashua, NH: www.alacron.com) had the hardware to interface to a new model of
high-speed camera interconnection technology called Camera Link. Intrigued,
Alacron arranged to borrow a new 200-500 Mbytes/s multi-tap Camera Link camera
to test its interfacing and other capabilities. Close evaluation revealed that
this new image sensor and interconnection technology rapidly would create
enormous improvements in machine-vision performance, as well as enormous
challenges to the machine vision-industry. Not only is the physical interface
to the camera vastly superior to current commercially available native PC
technology, but also its transmission rate meets or exceeds the storage and
processing capabilities of current technology. Channel Link technology is
spreading and probably will become the primary camera interconnection
technology in the near future. To do this, the primary historical factors
influencing the choices of manufacturers and consumers of machine-vision
technology must be understood and how the recent history of PC, data
transmission, and image-sensor technology has influenced these factors.
Driving factors
Users buy cameras to meet specific needs and speed requirements and
consider system price and ease of integration as primary factors in making
their choices. Similar factors influence systems integrators, who have to
provide application solutions at competitive prices. Thus, the kind of
interface implemented is not usually as important to the camera user as is the
price, the availability of many vendors, the available suitable technology, and
the ease of integration. Camera manufacturers design cameras with a view to
keeping the cost, power, and size to a minimum for a required performance
range. Additionally, they consider the availability of frame grabbers for their
camera, if one is needed, and either adapt their designs of their camera
interfaces to a commonly available one, or promote a specific frame-grabber
manufacturer who supplies a compatible interface. Frame-grabber manufacturers
are similarly interested in the lowest cost and broadest supported camera
solutions possible for a particular design.
Brief background
Prior to the mid-to-late 1980s, most machine-vision systems
were custom products produced as an integrated package that
was sold to a systems integrator or end user. Beginning with
the widespread adoption of the PC, manufacturers began to take
advantage of the widespread PC/AT-based computers and operating
systems to make modular merchant cameras and frame grabbers.
In the mid 1990s, the technological basis for the PC-based machine-vision
industry accelerated with the introduction of the PCI bus standard,
making available for the first time a native PC-based real-time
imaging system using widely available and multi-sourced cameras
and frame grabbers.
Until recently, analog cameras provided the standard interface. This
allowed customers, camera vendors, and frame-grabber vendors to design products
that met most everyone’s needs. Competition was based mostly on price and
features. With analog cameras as the standard interface, the main
differentiator was whether processing was done real-time on the frame grabber
or the data were transferred in non-real-time to the PC via the AT bus for
storage or future processing. Here, competition was at the frame-grabber
level.
Concurrently with the introduction of PCI technology, digital cameras
based on TTL, PECL, and LVDS emerged. These cameras yielded higher performance
for the camera manufacturers who could take advantage of the digital revolution
to support faster and larger formats. Camera manufacturers were able to meet
the higher frame and data rates, but now have to worry about compatibility with
as many framer grabbers as possible. Frame-grabber vendors now have to design
for the numerous, slightly different, digital interfaces that camera vendors
are providing, as well as to convince integrators and consumers of their claim
to the most ease of integration while charging the lowest prices. Therefore,
frame-grabber manufacturers further differentiated themselves into supporters
of native PC (for example, Bitflow, Imagenation, Integral Technologies, etc.)
versus co-processor groups (for example, Alacron, Coreco, Datacube, Matrox,
etc.), even further complicating the commercial market. These common or classic
camera interfaces and frame-grabber formats are summarized in Table 1.
Table
1. Common camera and frame-grabber characteristics
| Technique |
Data rate |
Availability
|
Standard interface |
Transmitter cost (camera) |
Receiver cost (frame grabber) |
| Analog Color
NTSC, PAL |
4-6 MHz |
Largest |
Yes |
$1.00 |
$12.00 |
| Analog BW
RS-170 |
4-6 MHz |
Large |
Yes |
$1.00 |
$7.00 |
| Parallel (RS-422) |
1-100 Mbytes/s
|
Medium |
No |
$20.00 |
$20.00
|
| Parallel (LVDS) |
40-320 Mbytes/s
|
Medium |
No |
$20.00 |
$20.00 |
Since 2000, further standardization of peripheral
buses (spurred by OEM PC manufacturers to enable the public
to easily add peripherals--or even cameras--to their PCs without
expert assistance) has had an invigorating effect on the choice
of camera interfaces. This has been further accelerated by the
rapidly increasing CPU and memory data rates commercially available
at competitive prices, which present an increasing option of
native PC-based real or near real time processing at speeds
only recently possible (see Table 2)
Table 1. Latest camera and frame-grabber characteristics
|
Technique
|
Data rate
|
Availability
|
Standard interface
|
Cost to camera
|
Cost to frame grabber
|
|
Serial
USB 1.0
|
1.5 Mbytes/s
|
Widespread
|
Yes
|
$10.00
|
Not needed
|
|
Serial
FireWire
|
50 Mbytes/s
|
Limited
|
Yes
|
$20.00
|
Not needed
|
|
Serial
USB 2.0
|
60 Mbytes/s
|
Limited
|
Yes
|
$10.00
|
Not needed
|
|
Camera Link
|
10 – 297 Mbytes/per channel
|
Medium
|
Yes
|
$4.00
|
$4.00
|
|
|
|
|
|
|
|
The ubiquitous USB 1.0 interface is now used for
low-speed, low-resolution cameras connected directly to the
PC, therefore eliminating the need for a frame grabber. However,
the limited distances, speeds, and resolutions of USB 1.0-cameras
severely limit their applications in a serious machine-vision
environment. The high-technology industry expects that USB 2.0
interface will supplant the USB 1.0 platforms as PCs evolve
their USB designs, but USB 2.0 may take time to become widespread
in the motherboard market.
The IEEE 1394 (FireWire) and newer USB 2.0 standards are fast enough for the
original analog camera signal in digital formats as well as
for low-to-medium speed, existing, digital cameras. Impeding
IEEE 1394’s utility, however, is the lack of widespread OEM
PC support for it in the industrial PC market (as opposed to
the consumer portable/Apple environment).The USB 2.0 interface
can provide higher data rates, but the same difficulty applies
with the additional problem that IEEE 1394 has wider current
acceptance. Even with the newest PCI, CPU, and memory technology,
the standard PC may have significant problems doing near real-time
processing without adding extra processors. After two processors,
the cost and shared memory contention potentially make a co-processor
(accelerated) frame-grabber substantially cheaper and more efficient
than the native PC solution. The advantage of using shrink-wrapped
software packages is also lost.
Camera Link emerges
The Camera Link standard is based on National Semiconductor’s inexpensive
Channel Link technology, which boasts superior transmission
rates of up to 2.38 Gbits/s (297.5 Mbytes/s) per channel. With
a standard connector and pin-outs, Camera Link, as currently
defined, uses one to three channels (Base, Medium, or Full configurations,
with a recent extension proposed by Basler Vision Technologies),
depending on video data format and rate. The Full configuration
delivers impressive transmission rates of up to 680 Mbytes/s
at 66 MHz or 850 Mbytes/s at 85 MHz. Moreover, this technology
is highly scalable and limited only by the number of chipsets
that frame-grabber makers put on a laminate or the number of
frame grabbers an integrator or end user is willing to put in
a chassis. For low- and medium-end cameras, frame-grabber manufacturers
and end users are back in the same competitive environment as
they occupied during the analog era. Camera vendors like the
ease of implementing Camera Link, and the low power and low
cost of the Channel Link chipsets. Frame-grabber vendors like
the new standard because they don’t have to do a lot of hardware
design, and the interface is easy to implement. Therefore, IEEE
1394 and USB 2.0 boards, and even single Camera Link boards
costing less than $50, will satisfy most native PC processing
needs. Customers like the ease of systems integration and the
low cost. So, it’s pretty much an all-around win.
In the last year, however, semiconductor performance improvements have
begun to expand beyond microprocessors and memory to CCD and CMOS sensors. This
trend will continue. For example, the performance of high-end CMOS imagers such
as Micron Imaging’s (formerly Photobit) MV13 (10 bits x 10 taps x 66 MHz) is
833 Mbytes/s and the MV40 (10 bits x16 taps x 66 MHz) is 1333 Mbytes/s. All
this data need a scheme for storage or processing that will tax the most
powerful current multiprocessor and envisioned uni-processor system in the near
future. This also includes accelerated frame-grabber manufacturers who bypass
the 64-bit/66-MHz PCI bus, as well as CPU memory performance issues.
For the foreseeable future, Camera Link will remain the only
standard digital ca mera at present, camera manufacturers are
acceding to the high-speed data acquisition markets and customers
(for example, military, semiconductor, and PCB inspection areas--which
constitute a non-trivial segment of the vision market) and are
beginning to manufacture cameras using the near 1-Gbyte/s rates.
However, such manufacturers will have to convince frame-grabber
manufacturers to support their cameras, and the frame grabbers
will require acceleration for use in anything approaching real-time
in the machine-vision market. Otherwise, such camera manufacturers
would quickly have to become makers of “intelligent” cameras,
which can perform significant processing and data reduction
before leaving the camera. Such an undertaking would require
significant digital hardware design and manufacturing experience
and a significant software effort, all of which most manufacturers
have been unwilling to undertake in the past.era interface with
the potential high performance and scalability needed by medium-
and high-end cameras, and the low cost needed by the low end,
of the industrial digital camera market. Given the past reluctance
of most camera manufacturers to expand to processing cameras
(and the required extensive software suites), the frame-grabber
vendors, systems integrators, and end-users will split into
two groups: a native PC, low-cost group, keeping pace with the
changes in standard PC technology and boxed software, and the
remaining users, who will require accelerated frame grabbers
or high-end multiprocessor servers with sophisticated data I/O
channels.With respect to either group, the continued adoption
and enhancement of the Camera Link standard is assured.