RDMA over Converged Ethernet technology enables direct memory access from the device to the computer.
BOB MCCURRACH, ASSOCIATION FOR ADVANCING AUTOMATION (A3)
The technical specifications of cutting-edge image sensors and state-of-the-art smart cameras mean absolutely nothing without the ability to get vast amounts of high-speed image data from point A to point B quickly, smoothly, and without any loss of fidelity.
Certain vision-guided robot applications require real-time feedback, which GigE (Gigabit Ethernet) Vision 3.0 can help deliver. Courtesy of iStock/kynny.
If the data cannot get to where it needs to go at a quality that is high enough for accurate calculation, integrators and engineers may as well depend on analog techniques to capture and analyze images.
If image cameras and sensors cannot communicate smoothly with frame grabbers or industrial computers, modern machine vision applications fall apart. Hardware manufacturers cannot solve these problems alone. It takes the entire machine vision community, working through organizations such as the Association for Advancing Automation (A3) and its technical committees to develop standardized solutions that empower the best use of machine
vision technology.
Currently, by leveraging RDMA over Converged Ethernet (RoCE) technology, A3 is setting the stage for another innovation that stands to influence the entire industry: the impending release of GigE (Gigabit Ethernet) Vision 3.0. The newest upgrade to the quintessential GigE Vision interface will make data transfer faster than ever before, enable superior error detection efficiency, and increase real-time processing and decision-making speeds.
History of GigE Vision
GigE Vision revolutionized machine vision by opening new opportunities. As machine vision applications increased in complexity from early use cases in the 1990s to more advanced use cases over time, demands for data throughput, latency, and stability grew accordingly.
The GigE Vision interface based on the Internet Protocol (IP) standard, developed by the Automated Imaging Association (AIA, now A3) and released to the public in May 2006, increased transfer bandwidths to 100 MB/s and data transfer lengths up to 100 m, exceeding the maximum usable lengths of FireWire, USB, and Camera Link. Increasing maximum cable lengths made it easier to deploy machine vision systems in difficult-to-access industrial environments, both on and off the
factory floor.
Hardware manufacturers historically provided proprietary drivers based primarily on USB 2.0 interfaces. Supported by most network adapter cards, GigE Vision became the new industry standard and interoperability between cameras and accessories increased dramatically. Almost any Ethernet-enabled PC or device became a viable option for machine vision system design.
Machine vision became more economical, flexible, and easier to scale. The new interoperability decreased peripheral development costs and passed the savings on to integrators. In many cases, the new interface eliminated the need for frame grabbers. Inventory costs decreased as GigE Vision supported industry-standard Cat 6 cabling replaceable from a multitude of suppliers.
Finally, A3 required manufacturers to complete a certification process before granting them a GigE Vision license. To gain certification, manufacturers needed to demonstrate that their equipment would be interoperable with any other piece of equipment that was also licensed to use the GigE Vision interface. The resulting, very wide range of interoperable hardware gave integrators unparalleled control over the acquisition, capture, and transfer of image data.
Evolving interfaces
The GigE Vision 2.0 standard, released in November 2011, improved support for 10 Gigabit Ethernet by augmenting clock speeds on cables. Now, a camera supporting GigE Vision could use more than one data transfer cable and increase throughput, boosting connections to faster image sensors and enhancing vision systems based on high-speed image capture.
A machine vision system inspectsa smartphone. Courtesy of asharkyu/Shutterstock.com.
The updated standard included an “all-in transmission” function to transmit full images using single packets, which is highly beneficial for applications running CMOS sensors that image small regions of interest at high rates of speed. Line-scan applications also benefited from the new transmission mode.
The final major upgrade introduced by version 2.0 leveraged Precision Time Protocol (PTP), allowing GigE Vision to synchronize clocks on each camera in a network with accuracy measured in microseconds. Adding 64-bit timestamps to images meant applications could correlate data from more than one camera. Multicamera vision systems installed in sports arenas, intelligent transportation systems, and machine vision inspection represent just a few applications that particularly benefit from PTP support.
All told, GigE Vision 2.0 optimized the synchronization between vision and non-vision hardware — such as lighting — and greatly served applications that require transmission of compressed images, and further minimized or eliminated jitter and other non-sync errors.
In August 2018, GigE Vision updated to version 2.1, enabling multipart payloads, or the ability to place different types of data into a single frame. Multipart transmission capabilities benefitted 3D imaging and other applications involving three-coordinate data structures.
Previously, cameras could use GigE Vision to transmit 3D data but only by using closed, proprietary systems that loaded RGB pixel data with 3D coordinates for transmission. The new functionality further cemented GigE Vision as a universal standard by once again eliminating proprietary systems and easing interoperability.
GigE Vision version 2.2, released in June 2022, offered generic data container (GenDC) streaming, meaning 1D, 2D, 3D, and multispectral image data, as well as metadata, can share a common data container.
RoCEv2 a precursor
The RoCE protocol, first specified in 2010 and used in large-scale deployments in 2013, allows one device to directly access the memory of another device without first passing through
an operating system, bypassing
the CPU to decrease load. The
protocol was developed to support data-intensive applications, such as cloud computing, Web 2.0 big data, and financial services.
A machine vision camera mounted to a multifunctional LED light inspects a car rim. Courtesy of Smart Vision Lights.
RoCEv2, announced in 2014 and specified in 2018, takes advantage of modern network interface card (NIC) technologies to refine and enhance the RoCE protocol for improved efficiency. RoCEv2 makes the protocol even more suitable for hyperscale data networks or enterprise data centers.
Now, working in conjunction with hardware developers, A3 will release GigE Vision 3.0, which integrates RoCEv2 and opens the door to the most advanced, flexible, and efficient machine vision applications yet.
Zero copy image transfer
The RoCEv2 protocol provides “zero copy image transfer” benefits to machine vision applications: Image data does not copy from source to operating system memory and then again to user buffer. Fewer steps allow for more efficient data transfer. Augmented and mixed reality applications, especially when used in collaborative settings, could make good use of the lower latencies provided by RoCEv2.
RoCEv2 also delegates error detection and recovery tasks to dedicated hardware, further increasing data transmission stability by increasing error detection efficiency. Medical imaging applications with virtually no tolerance for lost data will leverage this functionality to provide consistent results.
By integrating GigE Vision with RoCEv2, GigE Vision 3.0 generally frees system resources for all-important image processing tasks, a critical improvement for robotics or automation applications requiring real-time data processing for quick decision-making.
RoCEv2-compatible NICs include a range of links from 10 to 400 Gbps, which means GigE Vision 3.0 also provides increased flexibility and interoperability through better integration with Ethernet networks.
GigE Vision 3.0 enhances quality
The GigE Vision Technical Committee said that it intends to introduce GigE Vision 3.0 as a supplement, not a replacement of version 2.2. GigE Vision 3.0 will continue to use standard GigE Vision protocols for device discovery and configuration, ensuring continued support for all devices operating on earlier versions.
A3 plans to release the GigE Vision 3.0 update at some point in 2025.
The GigE Vision Technical Committee invites new members to contribute to the development of the new interface.
About the author
Bob McCurrach has more than 25 years of experience in the manufacturing and product development sector, with a focus on global engineering management, strategic programming, and process management. He holds a Bachelor of Science in mechanical engineering from Lehigh University and an MBA from the Georgia Institute of Technology. In 2011, he joined the Association for Advancing Automation (A3) as director of standards development for vision and imaging; email: [email protected].