How GigE Vision 3.0 Fits in the Machine Vision Ecosystem

Machine vision applications often benefit from cameras with higher data rates, which can enhance inspection precision, speed, or both. However, the limiting factor of these high data-rate machine vision cameras is frequently the cost. Generating, transporting, and processing large volumes of data can be expensive. Over time, the definition of “large” data volumes changes due to technological evolution. Faster computers and advancements in sensor technology continually push the price-performance ratio downward, making sophisticated machine vision systems more accessible.

 Transferring Data from Camera to Computer

Frame grabbers have been a staple in the machine vision landscape since the era of analog cameras. They excel at transferring data with minimal CPU usage. Over time, cameras have evolved to use standard computer interfaces like USB or Ethernet, which reduce system costs by eliminating the need for a frame grabber. However, they come with different compromises between flexibility, cable length, speed, and CPU usage. The frame grabber also offers the ability to perform image pre-processing functions, such as Bayer to RGB conversion, while transferring data without increasing CPU usage. This is something that standard interfaces cannot offer.

Related: Extending the Benefits of CoaXPress with CoaXPress over Fiber

 The Introduction of GigE Vision

When the GigE Vision interface standard was introduced in 2006, it allowed machine vision applications to use Ethernet as a transport layer for image acquisition without the need for a frame grabber. The standard, which the Association for Advancing Automation (Ann Arbor, MI, USA) hosts, emphasizes flexibility and interoperability between applications and devices. Performance and reliability depend largely on the quality of implementation. While GigE Vision defines mechanisms to achieve highly reliable image transmission, it does not specify performance requirements. As a result, there are significant differences in reliability depending on the devices or software implementations used in your application. These differences become even more pronounced in systems with multiple cameras or those employing high-speed cameras.

Since not all implementations are the same, it’s important to work with a vendor who has proven implementations of GigE Vision and can demonstrate no incomplete or skipped images. This is an absolute requirement when you are working with medical imaging applications or a multi-camera tracking system.

Related: How the GigE Vision Interface Serves Multi-Camera Applications

 The Challenge of Higher Speeds

Ethernet technology allows for transfer speeds that exceed the needs of most machine vision applications. Despite the availability of high Ethernet speeds such as 25 to 800 GbE, most cameras with transfer rates above 10 Gbps rely on frame grabber-based transfer technology. While the current version of GigE Vision allows for increased speeds, the cost of reliable solutions also rises as larger and faster memory is needed to manage packet retransmission. This limits the use of GigE Vision for faster Ethernet speeds. To address this issue, packet validation and retransmission need to be performed by hardware rather than the software stack. Historically, machine vision applications have relied on frame grabbers for this hardware offload. However, in recent years, network interface cards (NICs) from major brands have started implementing offload engines. The GigE Vision committee has been monitoring various technologies, and one protocol seems suitable for machine vision applications.

Related: What's New in CLHS Version 1.2

 The Emergence of RoCE

 RoCE (RDMA over Converged Ethernet) is a protocol that was initially designed for high-performance computing and allows direct memory access from a device to a computer without the operating system’s involvement. RoCEv2, the updated version, improves efficiency and flexibility by operating on the UDP/IP protocol stack, ensuring better integration and functionality within Ethernet networks. Recently, NICs have been developed with support for RoCEv2, which provides low CPU usage data transfer along with error detection and correction. RoCEv2-capable NICs are available in products featuring 10 GbE and higher bandwidth. Although RoCEv2 is not specifically targeted at image transfer, it is very versatile and offers multiple functions (verbs) for data transfer, each with its strengths. However, RoCEv2 does not define how to use these verbs to create a solution. This means two camera vendors using RoCEv2 can select different methods for performing the same task. The goal of GigE Vision is to bring interoperability between devices and applications for machine vision applications. This is why the GigE Vision Committee is working to release GigE Vision 3.0, which will bring interoperability for applications and devices using RoCEv2.

The current version of the GigE Vision specification includes three protocols: control, messaging, and UDP streaming. GigE Vision 3.0 will add a new streaming protocol leveraging RoCEv2 verbs. The control and messaging protocols will remain unchanged, with only new bootstrap registers introduced to configure the new functionalities. The current streaming protocol based on UDP will coexist with the new streaming protocol based on RoCEv2.

Related: GenICam: In the Engine Room of Machine Vision

 Conclusion

The evolution of machine vision standards continues to push the boundaries of what is possible in terms of data transfer, processing speed, and system cost-effectiveness. As new technologies like RoCEv2 are integrated into the GigE Vision standard, the future looks promising for more robust, high-speed, and interoperable machine vision solutions. This continuous improvement in standards and protocols will undoubtedly contribute to advancements in various industries relying on machine vision, from manufacturing to healthcare, ensuring higher precision and efficiency in their operations.