As Gardasoft celebrates 20 years of producing dedicated lighting controllers, we present the first in a series that reviews the history of machine vision and explores thrilling possibilities for the future.
Muffins and springs
Machine vision evolved from experimental systems that were first created in the late 1970s. The automotive
industry and its supply chain were some of the early adopters of machine vision and General Motors used imaging to check that collet holding springs were properly seated long before vision systems could be bought “off the shelf”. The components making up these early systems would be familiar to us now but in those days each individual item would have been sourced from a different application.
Just as today, the first vision systems consisted of a camera, a lens and an illumination device which functioned together to produce an image of the item to be inspected. The image would then undergo some kind of digital image processing or analysis to produce the required information. Minicomputers, such as the Digital Equipment Corporation’s PDP series, or even custom-designed hardware, were often used to process image data. The early cameras were always TV-type until smart cameras emerged in the mid-1980s. Smart cameras, which feature onboard image processing, eliminate the need for external computer processing so that just a result signal is transmitted back to the factory control system. At that time, if the ambient illumination was insufficient for the application, light sources such as quartz halogen or even fluorescent lights were used. Every machine vision systems would have been custom-built for its specific application and systems were very expensive. Yet, despite their primitive construction, these systems produced some very impressive results. One early industrial system was developed for muffin inspection and featured an image resolution of just 30 x 32 pixels. However, it proved highly effective at rejecting rogue, oversized muffins that would otherwise have jammed up the packaging machines.
Chips and processors
The capabilities of machine vision have been transformed since the 1970s by ever-increasing computing power, memory capacity and image sensor performance. Improvements in PC programming power and memory paved the way for the first PC-based image-processing toolkits and libraries. From these came the very first self-contained machine-vision applications, which had a simple interface framework and offered a plug-and-play approach for PC-based systems. The introduction of the PCI bus in 1993 enabled image data to be transferred within a PC and the advent of Windows 95 in 1995 made ‘point and click’ programming easier to implement. Image-processing boards with integrated processors were also developed to take advantage of the new Field Programmable Gate Arrays (FPGAs).
The advance of the PC clearly had a huge influence on image handling and processing capabilities, but the development of new semiconductor fabrication methods was also vital because of the creation of new generations of CCD and CMOS image sensors. By the end of the 20th century, a machine vision system would typically feature analogue output cameras using TV standards (eg NTSC, EIA, CCIR, PAL), frame grabbers with relatively long cables (up to 30m), Pentium II PC with PCI bus, and lighting consisting of halogen, discharge or fluorescent components. However, the 21st century was to herald a massive change in machine-vision technology and by 2010, a typical system would be quite different.
21st century revolution
During the first decade of the 21st century, new camera technology yielded more sophisticated cameras with vastly improved frame or line rates, resolution and form factor. The faster frame and line rates facilitated higher-speed inspections and made it possible to achieve multi-light inspections at a single camera station. Successive frames or lines could now be rapidly captured with different light configurations. The smaller camera form factors made it far easier to integrate cameras into the industrial process and integrate machine vision systems into crowded production lines. Further advances in microprocessors generated a huge rise in processing speeds while computing costs fell.
This advance in camera and computing capability was accompanied by the advent of highly-sophisticated image-processing software which offered an extraordinarily versatile array of tools for image analysis. Systems became used routinely for quality assessment, metrology, error/fault detection, sorting and process control. Product tracking and traceability was enabled by dedicated software for 1D and 2D code reading, pattern matching and optical character recognition. Simplified user interfaces with ‘point and click’ or ‘drag and drop’ capability made vision much more accessible to non-specialists. The advent of dedicated machine-vision data transmission standards simplified connectivity in machine vision systems and made components exchange easy.
New possibilities brought to light
A major breakthrough at this time came from research into super-bright LEDs. Experimentation with different semiconductor materials and phosphor coatings yielded lights with much higher intensity than was previously possible and a choice of wavelengths. LEDs offer very significant benefits over older illumination methods because of their much longer operational lifetime (typically up to 50,000 hours), better stability and low cost. The small size of LED lights allows them to be arranged in various geometries and it has become possible to create many different lighting configurations such as front and back lighting, on axis, diffuse, bright field and dark field lighting. The ability to rapidly pulse LEDs has made them the first choice for illumination in most machine-vision applications from this point forward.
In 2000, Gardasoft introduced the world’s first dedicated lighting controller for machine vision. It was now possible to precisely control the drive current to LED lighting and prevent illumination intensity from fluctuating. The new lighting controllers made pulsing, or strobing, the light easy to achieve and also allowed lights to be briefly driven at much higher intensity than is possible for continuous lighting. Pulsing the light also extends the lifetime of the light and can facilitate high-speed inspections. The new, dedicated machine vision timing controllers at this time also created the flexibility to sequence cameras, lighting and other components in a wide range of lighting and timing schemes.
Author: Jools Hudson, Gardasoft Vision Ltd, Swavesey, Cambridge, UK.
T: +44 1954 234970