In early 2003, soon after the first products for high-performance delivery of imaging data over gigabit Ethernet (GigE) came on the market, a handful of vision industry executives formed a committee to standardize GigE for machine vision. The goal was to create a framework for camera-to-PC communications so hardware and software from different vendors could interoperate seamlessly over standard GigE connections.
In May 2006, with the standard under the stewardship of the Automated Imaging Assn. (AIA), the first version of GigE was ratified. Today, more than 45 companies are committee members.
“It’s important to note that, from day one, the GigE Vision framework was designed to support a range of different performance levels and feature sets,” says Alain Rivard, vice president of R&D at Pleora Technologies. “This approach was taken so vendors would have room to innovate and differentiate their products, and vision system designers and integrators would have the broadest possible selection of compliant products with which to work.”
GigE Vision Version 1.0 has four main elements:
- Device Discovery defines how compliant devices such as cameras obtain IP addresses and are identified on the network
- GigE Vision Control Protocol (GVCP) defines how to specify stream channels, and control and configure compliant devices
- GigE Vision Stream Protocol (GVSP) defines how images are packetized, and provides mechanisms for cameras to send image data and other information to host computers
- An extensible mark-up language (XML) description file, essentially a computer-readable datasheet, maps the high-level features of a compliant camera to its low-level internal registers. This XML file describes seven mandatory features, and the descriptions must be presented in a syntax defined in the European Machine Vision Assn. (EMVA) GenICam standard.
These seven required features were chosen by the standard committee because they were considered the minimum feature set needed to support continuous acquisition and display of images from the simplest of GigE Vision-compliant cameras.
When adding other features, the GigE Vision standard recommends camera vendors use feature names set out in a separate reference document called the Standard Features Naming Convention. This avoids mind-numbing complications when, for example, camera vendors offer a brightness feature—one names it “Brightness,” while another names it “britenez,” and so on. Standard feature names are critical to achieving end-to-end interoperability.
The GenICam initiative was launched in late 2004 and its work has proceeded in parallel with GigE Vision. Its primary objective is to provide a generic camera control interface for all camera types, regardless of the interface technology they use or the features they implement, reports Rivard.
These days, networking is ubiquitous. GigE Vision leverages the widespread investment in networking technologies over the last decades—made outside of the machine vision market—by using those components, and specifying a protocol adapted to camera control and real-time transmission of images and video. “The main benefit to using Ethernet for machine vision is certainly the opportunity to use long cables—up to 100 m for copper—with a digital camera,” says Eric Carey, smart products manager, Dalsa Digital Imaging. “This is the first time analog cameras using long shielded cables can be replaced using a standard digital interface. All other technologies such as Firewire, USB and CameraLink can’t accommodate lengths more than 10 m over copper.”
Images are transferred as Ethernet packets, and protected by a checksum, whch ensures that if a transmission error occurs, the application can ask for retransmission. “The available bandwidth of Gigabit Ethernet also is perfectly suited to a majority of image processing applications,” adds Carey.
GigE Vision offers advantages beyond cable lengths for machine vision, says Dwayne Crawford, product manager at Matrox. “The GigE Vision standard surpasses the image fidelity of analog and the vendor interchangeability of Camera Link and LVDS/RS-422.”
Since machine vision involves connecting a camera to a computer, there’s no shortage of suitable applications.
“In fact, any application with low-to-moderate bandwidth requirements is a good candidate for GigE Vision,” says Crawford. “Camera Link still beats out GigE Vision in terms of maximum bandwidth.”
Less than a year after GigE Vision’s ratification by AIA in May 2006, you can already find many machine vision companies offering products supporting the specification.
GigE Vision is expected to improve interoperability, reduce cost and complexity in machine vision systems, and foster overall market growth. “However, it’s important to keep in mind that the standard by itself can’t guarantee reliability or performance,” cautions Rivard. “All GigE Vision-compliant products must be thoroughly evaluated to make sure they can handle the requirements of the application at hand.”
GigE isn’t the first standard bus to take hold in the machine vision market. “IEEE 1394 is one of the most popular buses for vision applications that don’t require high bandwidth,” says Matt Slaughter, product marketing engineer at NI Vision. “A standard PC bus reduces cost of cables and acquisition cards, and the large selection of cameras gives more flexibility in a vision system designed with IEEE 1394.”
Slaughter adds the biggest downside to IEEE 1394-based applications is that cameras have to be relatively close to the host. “If the camera is more than a couple of meters from the camera, the system engineer needs to add repeaters or hubs, which can cause noise problems,” he adds.
Since GigE cameras connect via RJ-45 connectors and use standard Ethernet cable, the cost of cables over these distances is much more affordable and easier to set up than other available options, says Slaughter. “Currently, the camera availability for GigE vision is nowhere near as large as that for more standard buses in machine vision, such as IEEE 1394 and CameraLink, but as the bus takes hold in the market place, this limitation will decrease as well.”
Another important consideration is the amount of jitter that can be tolerated. Most machine vision applications require a maximum guaranteed response time, and determinism is a measure of confidence that the event will be completed in the required time. As an application’s guaranteed response time decreases and the required level of determinism increases, its tolerance for interruptions and jitter decreases.
“Since Gigabit Ethernet’s architecture, in other words, the presence of routers and hubs, device sharing, packet reconstruction on the host and location of auxiliary I/O, can affect an application’s jitter, the point-to-point, non-packet-based protocols—such as Camera Link—might be more suitable,” concludes Crawford.