GigE: Vision Industry Gets Ahead of the Tech Curve

Underdogs are great. They go up against the 'big guys' and sometimes they win, even with the cards stacked against them.

In the world of high tech, machine vision has welcomed the underdog role. Although it did not drive the development of many of the computing and imaging technologies used in machine vision, the machine vision industry benefited from scientific and defense imaging systems that led to the commercialization of solid state imagers such as charge coupled device (CCD) cameras. The demand for personal computers led to advances in complementary metal oxide semiconductors (CMOS) that resulted in new classes of imagers and general purpose microprocessors instead of expensive ASICs. Finally, computer network technologies emerged, such as Ethernet. Again, machine vision benefited from a consumer technology that it adapted for industrial purposes.

All great tools, but none exactly what the machine vision industry wanted or needed. The situation was further complicated by the vision industry itself. In a group of underdogs, it can be hard to create consensus that carries sufficient weight to sway developing technology.

Today that is changing. Machine vision companies still lack the resources of the Microsofts and Intels of the world, but the industry is learning that it is better to be a pack than a lone wolf. Perhaps no where is there a better example of today's stronger, smarter machine vision industry than the industry's efforts surrounding the emerging Gigabit Ethernet Vision (GigE Vision) and GenCam standards.

When high speed isn't enough

Past advances in machine vision have often focused on faster processors, faster algorithms, faster read-out electronics and faster networking protocols, but faster isn't always the whole answer.

In recent years, the machine vision industry established the Camera Link® network standard for transmitting data between cameras and image processing platforms. A point-to-point network, Camera Link offers better bandwidth and transmission length than consumer-based USB and FireWire standards. Yet it still lacks the distance needed for many emerging vision applications and the networking flexibility to support distributed processing platforms. Camera Link transmits 255 MB/s with one connector and up to 680 MB/s with two connectors across distances of up to 10 m. (see related AIA story).

Moreover, Camera Link is still a specialized technology. Its chipsets are not embedded in every PC motherboard as Ethernet chipsets are today, hence the need for framegrabber boards. ‘‘When you have a high-bandwidth imaging system, you still have to get the data into memory somehow,’‘ said John Merva, president of GigE camera maker, Tattile USA (Bedford, New Hampshire). ‘‘With standards like Ethernet, you can buy a $60 NIC (network interface card/chip) and inexpensive cables instead of framegrabbers costing several hundred dollars, and expensive Camera Link cables.

‘‘Now some framegrabbers do more than just port data into memory, but not all applications require that,’‘ Merva concluded.

Getting the standard before the technology

In 2003, George Chamberlain, president of GigE hardware and software developer Pleora Technologies (Kanata, Ontario, Canada), and Toshi Hori, then president (now CTO) of JAI PULNiX (Sunnyvale, California) and a proponent of vision standards and Camera Link, joined forces with other industry leaders to standardize how GigE is used by the vision industry.

‘‘We realized that GigE had the potential to satisfy an expanding variety of imaging applications,’‘ said Chamberlain. ‘‘We wanted to get a GigE vision standard in place before the technology started to take off, so that we didn't have a range of different implementations.’‘

Among GigE’s benefits for vision is its low cost, mass-produced chipsets, cables and associated hardware designed for the consumer industry. It is also flexible enough to accommodate virtually any camera-to-PC networking requirement, and has enough bandwidth to transport image and video data in real time.
 
In 2004, AIA agreed to assist the nascent GigE Vision standards group, using the lessons learned from earlier standards efforts.

‘‘Camera Link defines a serial control signal path, but doesn’t define the protocol, so each camera company does it differently,’‘ explains Chamberlain. ‘‘Some are binary, some are ASCII, some have check sums and some don’t… but the new GigE Vision and GenCam standards will provide one control interface that supports many different cameras across a common network platform.’‘

Moving forward

The first implementation of the GigE Vision standard is expected later this year. A collaborative effort of AIA's North American and European Working Groups, the standard includes five parts: defined bootstrap registers on the camera; an XML descriptor file that self-describes the camera's capabilities and command controls; discovery, control/command and video/data control protocols operating at layers 5-7 of the OSI protocol stack; and reference software for validating the standard. Once adopted, the standard will allow any application programming interface (API) to automatically discover, control and communicate with any GigE Vision-compatible camera.

The GenCam standard is a separate initiative from GigE Vision that provides a common software interface for any vision camera across all major network protocols, including FireWire, USB, Camera Link and GigE Vision. Its first implementation is also due out later this year.

Applications such as traffic monitoring, security and surveillance, and distributed machine vision in large factories, grocery stores and similar facilities require longer transmission distances at high bandwidth. High-resolution imaging requires more processing, but dedicated processors are expensive to develop, manufacture and support. Machine vision systems that can distribute images to clusters of commercial PCs can accomplish more for less compared to the old paradigm of dedicated vision processors.

Until now, the missing piece has been a controllable network that links all these nodes in real time. Standards such as GigE Vision will go a long way towards supporting those applications, while providing the scalability, flexibility and ease of use that industry wants. At the same time, vision vendors will be able to differentiate themselves by how they manipulate the massive amounts of data inside the PC so as to not overtax the CPU, and through the application of FPGA (field programmable gate array) and other devices to prepare the data for GigE transmission and limit end-to-end latency. Then of course there are all the usual suspects of algorithm efficiency, API, HMI, programming languages, etc.

The high tech world is constantly changing, but this time, the vision industry is ready.