Introduction to Machine Vision Standards
The range of interfaces available for modern image processing
How does an image in a camera get to the PC where it is evaluated? To this apparently simple question, the industrial image processing sector can provide a range of possible answers that are currently the object of heated discussion. You will hear terms such as "CameraLink", "IEEE 1394", also known as "FireWire" available as version FireWire A or FireWire B, together with "USB", "USB 2", "Gigabit Ethernet" or "10 Gigabit Ethernet". Alongside these standard transfer methods that primarily come from the consumer PC market, there are also a number of proprietary transfer technologies. This article provides an overview of available standards.
For an image to be processed, it must be converted into a digital format by the sensor and then sent to a processor, unchanged and undamaged, via the quickest possible route. The most common method used in modern image processing is to transfer the image data via an external line or a bus to the processing unit's main memory. One of the advantages of this procedure lies in the fact the processor can do its work in an ideal, cooled environment, for example in a PC, without its dissipated heat interfering with the sensor.
In particular in the field of PC-based image processing, the number of data transfer possibilities has increased greatly in the recent past: The market offers frame grabbers with PCI, PCI-X and PCIe connectors to acquire analog or digital signals like Analog, CameraLink, IEEE 1394 or FireWire - as version FireWire A or FireWire B - USB, USB 2.0 or Gigabit Ethernet. Alongside these standard transfer methods that partly come from the consumer PC market, there are also many other proprietary transfer technologies.
Three decisive criteria
Three properties are of primary importance to industrial users: the flexibility, availability and stability of the employed components.
A high level of flexibility, combined with a low level of risk in the employed image acquisition components, is especially important to mechanical engineers, since it is usually only possible to test the image processing system under real-life operating conditions once the machine has been completely assembled. If the production environment proves to be different from the laboratory test environment, it may be necessary to make extensive changes. If, in such cases, a higher resolution camera or a faster frame grabber is needed for example, it should be possible to implement the modification quickly and without any great effort.
The second decisive point for the industry is the long-term availability of the components. The time required to develop a special machine from planning through to realization is usually over a year. Such machines should then run for at least five years. The components should be available in their original form for at least this period because if an image processing component fails during the machine's lifetime and is no longer available, then the user - as well as the supplier - is faced with enormous difficulties.
Thirdly, component operation must be stable and guarantee the best possible level of data and reliability. Here again, costs are an important consideration: if a production system shuts down then this results in high costs. Consequently, a system must under no circumstances fail because of an image processing system. If the acquired images are damaged, or even worse, lost during transfer, this can have dramatic consequences for the manufacturer.
The most important criteria on which the selection of an image processing system should be based are not therefore the transfer technology or the associated interfaces only. Whenever a component is chosen for a system, the three above-mentioned criteria should always be considered as well.
A comparison of interfaces
Even if no end to the use of the most frequently employed analog technologies is expected for some time to come, it is worth while taking a look at the relatively recent interfaces that many vendors promote enthusiastically as the technologies of the future.
Of these, FireWire/IEEE1394 and USB have recently enjoyed great demand with their extensive product range from various vendors.
With the increasingly great range of CameraLink cameras, CameraLink is taking over from the previously common LVDS frame grabbers even though, here again, the choice of the transfer medium (LVDS or CameraLink) depends on the employed camera. Consequently, cameras with an LVDS interface will continue to exist but there won't be many new models.
Although Gigabit Ethernet is still in an early stage as far as industrial image processing is concerned, it plugs the gap between the lower end digital cameras on the one hand, and CameraLink on the other. What is more, Gigabit Ethernet is the only technology to come from a mass market that possesses both the innovative pressure and the financial resources to drive its development forwards.
In general, we can say that CameraLink will tend to be used for cameras with high transfer bandwidths. In contrast, IEEE1394 and USB are especially suitable for lower bandwidth applications. In this scenario Gigabit Ethernet builds the bridge as it almost triples the bandwidth of a FireWire A camera and offers long cables in addition, but can not reach the bandwidth CameraLink offers.
What, however, none of these technologies offers is real hardware abstraction. This means that as far as the application or the programmer are concerned, a universally employable software platform which can be used to address a very wide variety of data sources in a uniform way without being tied to a particular technology or particular hardware vendor, is again desirable.
Both forms of FireWire (IEEE 1394 A und B), USB 2.0, and Gigabit Ethernet offer users a major advantage compared to the conventional technology, primarily in that they permit simpler hardware integration since no frame grabber has to be installed in the computer, thus often making a considerably more economical cable concept possible. On the other hand, these technologies introduce a higher CPU load then what the system designers have been used to with frame grabbers. FireWire and USB also lack fully-fledged industrial cable concepts and the standardized connection of the components to the widest possible range of software platforms has not yet been achieved.
This is understandable when you consider that both FireWire A and USB 2.0 are consumer driven technologies adapted by the machine vision industry. The design premise of such interface technologies is much different from pure industrial technologies. FireWire and Gigabit Ethernet appeared in the portfolios of machine vision suppliers not long ago. Among these technologies Gigabit Ethernet is the most industry proven technology. It offers a wide bandwidth of industrial components which are necessary for a Gigabit Ethernet infrastructure like switches, cables etc. GigE Vision as an Ethernet protocol standard in combination with GenICam simplify the integration and exchange of hardware components and so come very close to a complete hardware abstraction.
Due to the number of products available on the market, the question of which technology is the right one in view of the task to be accomplished is often difficult to answer. It is therefore advisable to draw on the expertise of the EMVA members, which ideally, should offer the greatest possible number of components and possibilities, thus guaranteeing independence and objectivity
It is not just the transfer technology between the camera and image acquisition phase but the choice of the software and the right sensor technology that is decisive when putting together an image processing system. The sensor must fit the demands of the imaging task and the software must allow the user to employ the widest possible variety of hardware technologies in order to ensure flexibility and facilitate possible substitutions. When choosing the hardware, industrial users should look for proven, stable components that will continue to be available in the long term.