How Can A Camera Capture Very Subtle Differences Between Colors? Help

My client’s facility manufactures sunroofs for the automotive industry. Their factory is incredibly automated and uses multiple industrial robots for many of the manufacturing operations. They’ve asked me to come up with a video inspection & recording system to help them inspect the adhesive that is applied to the glass just prior to pressing the glass into place on the sunroof’s frame. A human inspector will be doing the inspection via a 1080P monitor. We’ve figured out the video recording elements of it so they’ll have a video record of each product during its inspection along with the product’s serial number.

The problem is obtaining a good image of the adhesive so the inspector can easily spot flaws (gaps or low areas mostly) in how the adhesive was applied. The adhesive is applied automatically by the robot holding the glass under an automated applicator (basically, a glue gun) which applies a bead of adhesive around the entire perimeter of the glass. The adhesive bead is about ½” wide with a triangular cross-section about ¾” high. The perimeter of the glass is painted a very dark flat gray (nearly black) – it’s painted about 3” wide all around the perimeter. A ¾” wide (roughly) stripe of black primer is applied onto the dark gray painted surface to help the adhesive do its job. My client is perfectly willing to program the robot so the glass (with the primer and adhesive applied) is repositioned about 3-4 feet in front of the camera and wait until the operator has viewed the monitor and is satisfied with the bead of adhesive. If OK, the robot (still holding the glass) moves it and presses it down into the frame before starting again on another piece of glass.

The problem is that everything is black or nearly black. The perimeter of the glass is a very dark flat gray, the primer is black (with a bit of gloss to it) and the adhesive is also black and is quite glossy. We started by using a 5MP IP camera from 3xLogic (VSX-5MP-D2) for the inspection and have experimented with a number of different cameras trying out cameras with WDR and IR. We also tried a B/W camera. And, of course, we've adjusted contrast, white levels, etc. Unfortunately, nothing we’ve tried can discern the difference in contrast between the blacks sufficiently for an inspector to spot any flaws. We’ve suggested using a different color adhesive and/or primer but that’s out of the question. For what it’s worth, they’ve been doing visual inspections for quite some time (i.e. the robot holds it up in front of an inspector) so there’s no problem with a human eye telling the difference. But we’ve yet to find a camera that can.

Sorry I can't include any photos or drawings -- my client's rules, not mine. Any suggestions would be very much appreciated.


Thanks for the very detailed explanation. I know what you mean. It's very hard to see such nuanced differences with camera default configuration.

One crazy suggestion. How about thermal? If the adhesive is a different temperature maybe that will show up on the video feed.

Let's see what others have to suggest.

Thanks, John. We thought about thermal -- the adhesive is slightly warm when applied -- but we rejected that because we felt the difference in temp was pretty small plus the heat from the adhesive would transfer to the primer pretty quickly making them the same temp pretty quickly.

How about you add a white LED illuminator or an IR illuminator? Would that better highlight any subtle differences in color?

There's a good chance the adhesive is more sensitive (absorbs) certain IR or UV wavelengths the glass doesn't. Then you can find a specific filter plate that screens everything else out.

Call the chemical adhesive supplier and talk to the chemist if they can help.

The USAF uses a similar process to confirm repairs on jet engine compressors. If cleaning chemicals leak out through minute cracks in a housing, filtered cameras detect it.

the heat from the adhesive would transfer to the primer pretty quickly making them the same temp pretty quickly.

The emmisivities of the two substances though would likely differ and therefore offer some additional contrast. Also, its not necessarily one or the other since Flir has released MSX, a hybrid visible/IR technology in some of its handhelds, though I not sure what their IP cameras plans are... Here's a link for it with a video about it, if you haven't already seen it...

Rukmini knows his stuff. I can't see the original application description but would suggest talking to a Flir Thermography or Scientific rep. Things like spectrum filters, mid or short wave imaging along with emissivity could create an image with extremely accurate sensitivity. With that comes a higher price though.

I’d suggest experimenting with your illumination source. Is there a part of the spectrum that shows up the contrast better than others? Perhaps a relatively tight band around red, green or blue? If so then maybe a monochrome imager with a spectral filter that corresponds to the illumination will work so you aren’t fighting against the camera’s integrated color filter.

I wouldn’t rule out the Near IR spectrum either. Finally take a look at off axis illumination and detection. On some targets the contrast can really “pop” when exposed to side lighting. Although this can make registration a bit tricky, but that doesn’t sound like priority for your application. Good luck.

Very interesting application.

Couple of suggestions.

  • Different lamp types and angles.
  • A good quality display that is capable of reproducing required contrast levels. Thinking of plasma for the best black level reproduction.

Interested in other opinions as well...

Maybe an SDI camera, like the Marshall Electronics CV-355-CS? Run RG6 back to a converter like the Blackmagic UltraStudio Thunderbolt or UltraStudio SDI, and plug that into a computer, and plug that computer into a high quality, color accurate monitor like the PA272W. You're going to need a really good lens, too.

Color accuracy is hard to do, and most surveillance cameras are not designed with color accuracy in mind, because color accuracy isn't really as important in a surveillance application.

This was my first thought as well.

You need to see subtle nuances in color and brightness, and the problem with almost all types of compression is that they inherently crush and throw out these differences - that's simply how they're designed to work. SDI would give you an uncompressed, unadulterated image.

The other thing you might want to do is set the camera's white balance manually, as AWB can be affected by large color areas in an image... case in point was a site complaining that a camera image was "really blue", but when I looked at it, the majority of the center of the frame was filled with a yellow patio umbrella that the AWB was trying to "compensate" for.

Hyperspectral imagers are designed to achieve much finer color gradation than tri-color cameras.

It would be interesting to see if a local university (or anyone with a hyperspectral imager) could help you determine if there's any distinguishable difference that stands out clearly in hyperspectral imagery.

Instead of a surveillance camera or camera company, try contacting machine vision camera makers. They may have more information or specialized cameras for those applications since it sounds more like a machine vision application.

You can try this contact if you want:

Nathan Cohen

Saber1 Technologies

(978) 244-0490 x 113

I've talked to him before about machine vision cameras and lens filters and he was pretty helpful, though surveillance concepts are kind of foreign to him. :)

If Cost is not an issue and performance is, go to Tektronics and Fluke industries.

Need Extremes (.oooo1 lux or better) Color Spectrometers, computer based systems, not video based systems.

The engineers there know the right people to get thermal imagers and High tech equipment to complete the task.

CCTV cameras wont do this job.

You need to go to thermal and digital imaging equipment (Industrial side of the industry)

Thank you all for your responses. Using a thermal camera such as those offered by Flir has been considered on our end, but I'll have to check with my client to see if his budget can handle it.

We've tried a bunch of different lighting types, angles, etc with barely discernable differences -- mostly we were trying to capture the reflectivity of the adhesive, but it's not consistently flat enough to count on the light reflection. Filtering the light is an interesting idea though.

I like the idea of an SDI camera. I'll have to look into that.

And I'll definitely call Nathan at Saber 1 about machine vision cameras.

Cost is always an issue for any firm and my client is no exception. But their way of thinking about ROI is quite different from other clients and we may be able to work something out.

Thank you all again!

I finally, using a PC, have been able to see the full question. This is absolutely in the realm of a machine vision / thermal camera. To give you an idea of the uses, there are box packaging systems that "hot glue" each container shut. the thermal camera "images" the box after the ends are folded and makes sure the area is both hot enough and large enough to have properly sealed the container, through the cardboard.

This is not a security camera application for thermal, it would require a machine vision version with much higher resolution and temperature accuracy.

How about looking at machine vision applications, that's more the ballpark you're entering. Some machine vision cams are good at this. Filters may also help, neutral density, perhaps a circular polarizer to eliminate the glare off the glass.

I would look into various type of scanners. Not sure what they have for Industrial type of situations, but they would be able to find small errors.

If you use a camera, you should create a lot of bright light on the backside of the glass to highlight the imperfections on the front side of the glass.