Color Machine Vision Goes Beyond the Visible
| By: Winn Hardin, Contributing Editor
The use of color cameras in machine vision applications used to be limited in part by limits to handing the amount of data a color camera could produce, particularly three-chip cameras compared to single-chip Bayer filter color cameras. But as computer processing power has increased and the number of commercial applications continues to grow, color machine vision applications have increased in volume and as a percentage of the overall machine vision market.
Flat-panel displays, adhesives, food processing, remote sensing, and print inspection are just some of the applications driving the use of color machine vision, but many of these applications are now looking beyond red, green, and blue color channels to the infrared (IR) and ultraviolet (UV) spectral bands to find defects and make measurements that cannot be made using only visible light, while other color systems use combinations of colored lights and filters with monochrome cameras to make cost-effective color measurements in specialized applications.
Growing Color Vision Market
“Color machine vision applications used to account for about 10% of the market, but we’ve seen that grow to about 15% as the cameras have improved and experience has grown,” notes David Gilblom, President of Alternative Vision Corporation (Tucson, Arizona). “We are now seeing many applications where color is essential actually being implemented.”
Color vision continues to be an important solution for print quality and color registration; food sorting where IR imaging is growing because of its ability to separate different organic materials and identify “bruised” food based on water concentrations; and airborne remote sensing applications for crop monitoring. While monochrome cameras paired with near infrared (NIR) bandpass filters have evaluated fruit for bruises for many years, some color machine vision manufacturers are now adding a fourth chip to a three-chip camera specifically for the IR band.
“JAI (San Jose, California) has two-chip cameras that go out to 900 nm for NIR sensing,” adds Gilblom. “We’ve worked with Dutch company Quest Innovations (Middenmeer, The Netherlands), which has built multispectral cameras with up to five chips for remote sensing applications. And if you look at the leading edge in broadcast applications, you can achieve the best color accuracy with up to about six channels, but the machine vision market doesn’t tolerate camera prices like that. Point-measurement colorimeter systems for medical monitors and other applications are relatively cheap; cameras are not.”
Imaging colorimeters, or camera-based systems that can make absolute color measurements independent of intrinsic system and extrinsic environmental factors, have long been sought. However, color response can be affected by so many factors, from ambient conditions to aging sensors and lights, that creating a bulletproof imaging colorimeter remains a difficult and expensive enterprise.
“The best accuracy for an imaging colorimeter is about 1 Delta E, while many colorimeter applications require a Delta E of 0.1 or 0.01,” says Gilblom. “You have to be very careful about calibrations, sensor and illumination uniformity, and making sure that any objects in the camera’s field of view are essentially the same color. It has to be very carefully controlled.”
Color and Light Uniformity
LED lighting manufactures are aware of the challenges facing imaging colorimeters and are developing solutions of their own.
“We recently had an application where a client wanted to check red writing on a red product and needed to make sure the red shading was within an acceptable range,” says Oliver Szeto, President of Metaphase Technologies Inc. (Bensalem, Pennsylvania). “By using the Metaphase’s new Universal Light Controller-2 (ULC-2) and its Ethernet control capability, we can monitor color changes in the images from the camera of a color calibration target in the field of view to make sure the light intensity was the same from shot to shot...We guarantee uniformity of +/-5% with our LED lights, but for one of our Print Inspection OEMs we recently provided +/-1% uniformity with a 45-inch LED line light using a combination of special driver electronics and advanced Optics.”
Uniformity is important for the hot application of LCD flat panel inspection. Using linear cameras and LED lights that can stretch up to 5 meters and beyond, LCD flat panel inspection demands high uniformity across the full length of the glass inspection system. Metaphase accomplishes this by putting the LED drivers right inside the LED light itself (a driver for every set of LED components that are located adjacent to each other) along with special lensing and diffusers that allows them to make lights of virtually any length without gaps that can result in ‘hot spots’ or hard reflections from the glass surface.
Kamier Fariba, Director of Software Engineering at eCullet and customer of frame grabber manufacturer Engineering Design Team, Inc. (EDT) (Beaverton, Oregon), uses a 36-inch white LED light to sort a very different type of glass: recycled bottles. Fariba’s system uses an EDT frame grabber and a JAI three-chip CCD camera to separate different pieces of crushed, falling glass. “Color selection was very important, so we do both offline and online calibration by tracking maximum intensity of individual pixels. Also, we had to go with a three-chip design because a Bayer filter camera would introduce too much spatial uncertainty when activating the downstream actuators – a bank of air jets about 12 inches past the camera. If we could only have the three linear arrays arranged to remove the distortion from using a single industrial lens, that would be great. There are optics out there that do not distort, but they’re very expensive.”
According to Alternative Vision’s Gilblom, the flat-panel industry also struggles with optical distortion across large areas. “As flat panels grow in size, the cameras and optics have to get longer, too. And it’s hard to make cost effective prisms that are 24,000 elements long so the shift in this market has been to multiline sensors.”
More Is Never Enough
LED chip manufacturers are also making it easier for machine vision light vendors to match LED light emission spectrum to the specific needs of an application.
“Traditionally, machine vision systems that use LED lighting would use red LEDs because they offered high output power, but red light doesn’t work for color applications,” notes Mark Kolvites, Technical Sales Manager at Metaphase. “Users would want to use white LEDs for color applications, but these LED components used to emit a very ‘bluish’ white light. Now, we can select white light LED chips from different color bins and make a machine vision light that exhibits a very pure white color, or a blue or yellow white light, depending on what the customer needs.”
Infrared lights, once the darlings of the telecom community, are being used more in industrial – and particular – traffic applications where imaging systems want bright illumination without posting a danger to motorists. With appropriate optics and IR LED illuminators, an IR sensitive camera can capture an image of license plates across 4 or more lanes of traffic. For high speed traffic applications, Metaphase strobed IR illuminators provide the high intensity and camera synchronization required to capture a non-blurred image necessary for car identification/license plate reading.
At the other end of the spectrum, Metaphases Szeto says his company is seeing increasing demand for UV LEDs down to 265 nm and below. “UV LED lifetimes used to be short – about 1000 hours, but we’re seeing 50,000 hours for 365 nm LEDs, and up to 10,000 hours for LEDS down to 255 nm. The LED manufactures are gaining more control over the doping and manufacturing processes for these chips, so they’re brighter and lasting longer.” The 395nm UV LEDs are more than 10 times brighter than previous versions
While color machine vision systems remain a relatively small part of the overall market, they can offer strong margins because the technology is typically an option of last resort. In other words, color machine vision can do what few other quality control systems can for many applications – and that demands a premium.