Public Lab is launching a new 🎈 Mini Balloon & Kite Kit as part of Kickstarter Gold -- take a look!

Public Lab Research note


  • 1

Red filters in Infragram Webcams: a test

by warren |

This test is nowhere near as comprehensive as the big one we did at LEAFFEST a few years ago, but I wanted to know if we could switch Infragram Webcams to red filters as we've begun to do with other Infragram conversions. I set up one of each and ran it through a couple test scenarios, but I'd like to ask folks input on if this seems to be working.

The first is of a bed of weeds in a nearby park. Running with the preset for blue filters, colorized NDVI:

2015-05-27T19_39_58.552Z.jpg

And same but with the red filtered camera, with the red filter preset:

2015-05-27T19_39_55.170Z.jpg

The original image is:

IMG_20150527_153042.jpg

Infragram links:


I did a second test of a potted houseplant on my back porch:

Blue filter:

2015-05-27T19_42_55.993Z.jpg

Red filter:

2015-05-27T19_42_59.348Z.jpg

Infragram links here:

Blue: http://infragram.org/i/556618eafed547da29000061 Red: http://infragram.org/i/55661939fed547da29000062


One thing I immediately notice is that for areas with low dynamic range data, in dark regions like shadows, the blue filter technique returns lots of artificially inflated values (red artifacts) which likely aren't really high productivity zones, but just noise. The red filter technique does not cause this effect.

It does seem like the red filter version can work with this webcam. But I'd like to hear from others to get consensus.



ndvi webcam infragram plants vegetation infra-red infrablue test infragram-webcam

response:10996

16 Comments

I'm confused about which images are photos from the camera and which are NDVI images. It would also help to know what the colors mean in NDVI images.


These are all NDVI except the third, normal looking one which I labelled "original". Unfortunately a bug in infragram is not allowing display of the unmodified raw images as its supposed to. The scale is the default for Infragram.org.


Can you attach the original photos from the camera? Or do they exist only locked in the burly depths of Infragram for eternity?


I think they're gone, honestly. The bug seems to affect images captured via the webcam direct connection, but not images which were uploaded. I don't have time to go out and recapture at the moment, though, as I have to be working on this Kickstarter project :-/


Do you think there's nothing you can infer without the originals? I'm guessing you want to look at the blue/red gap?

These are the camera serial numbers, btw-- I think these may be a new edition. tmp_12229-IMG_20150528_1230112096177160.jpg


Yes, I can't say much without seeing the originals. Someone familiar with what infragram.org does might be able to infer something. When I follow your link to infragram I don't see the color key. So the NDVI images provide very little information.


The colormap is based on one of yours: colormap.png


You're right - we need to display the colormap in the saved images too -- but we haven't had any time to do further development on Infragram.org, sadly.


It looks to me like both offer pretty good differentiation, but that the red filtered version depresses NDVI values. But since we're not assuming absolute intensity calibration here, the relative data is still useful -- but it might be a bit confusing if folks are looking specifically for positive NDVI.

It's too bad we don't have the originals to experiment with.

I wonder if there's some standard coefficient we could add to the red-filtered ones based on a more structured test, perhaps following Ned's absolute calibration methods. If the coefficient were consistent (i.e. if red filtering consistently reduces NDVI by a factor or a fixed amount), we could just add that as a variant of the red filtered preset, just for use with the webcams.


A standard coefficient is a tough nut. The amount you would want to shift the NDVI values will differ a lot among photos. Exposure and color balance both influence how an infragram photo will translate into NDVI. It is not just the difference between red and NIR, it is the absolute value of red and NIR, at least as that changes their ratio.
.
NDVI_diverge.JPG
Infragram NDVI is a function of the difference in brightness between the red and NIR values in a pixel. The bigger the difference, the bigger the NDVI value, all else being constant.
.
NDVI_parallel.JPG
NDVI is normalized by relating it to the absolute magnitudes of the red and NIR values, so dim pixels (left) and bright pixels (right) with the same absolute difference between red and NIR will have different NDVI values. This is why exposure is important-- bright photos of healthy leaves can produce very low NDVI values, and dim photos of the same leaves can produce high NDVI values. However, bright and dim leaves in the same scene are often illuminated by light with different color balance (e.g., direct sunlight is reddish) so this relationship is never simple.
.
Because everything that is done at infragram.org is subjective, there could be a slider that shifted the NDVI values into the color range of choice, and another one that spread those values across a wider range of the color scale. The values of those two adjustments could be displayed with every iteration of the result.


Hi, Chris - I've been plugged into several other projects and really didn't have time to do this properly, and there was some confusion over the webcam model that the Kits team sent me. If we sent this prototype red filtered webcam to you, would you be willing to do a more rigorous test?

I think you mean "relative" and not "subjective", right? Also, I'd prefer that people do the shifting by changing the color map, rather than changing the values. Or, at least that we choose one or the other, since they both do the same thing, but having two points where it's possible to do it is confusing.

Reading over Ned's procedure for automating absolute calibration, we should be able to add that quite easily to Infragram.org as well -- potentially even with the auto-recognition of a QR code. Exciting, if we ever have time to put some work into the Infragram.org codebase!


I think you're right that subjective is the wrong word. Maybe qualitative. A plant with an actual NDVI value of 0.5 may have an infragram NDVI value of 0.3. Two plants that actually differ by 0.4 NDVI values may differ by only 0.1 at infragram.org. Infragrammar allows users to adjust the results, so those results can be subjective.

I guess it's straightforward to port Ned's math to infragram.org, but following the entire calibration procedure is probably not for the typical infragram.org user.

  1. You need two targets of known reflectivity in the red and NIR
  2. The targets must be exposed well
  3. You have to tell the routine where the targets are in the photo

To do the correction for extraneous NIR in the red channel

  1. You have to know something about the filter you're using
  2. You have to know something about the sensor you're using
  3. You have to figure out if that correction actually works

Ned doesn't know yet if a camera with auto white balance will give good results. Ned's procedure will have to be tested on an infragram web cam.

I won't be able to do any tests for a few weeks (tundra beckons). All I need is some images of interesting test scenes with good (or bracketed) exposure, I don't need the camera. Maybe that is a good agenda item for LEAFFEST.


Specifically, I think we need to encourage people to structure experiments around what these cameras do best at the moment -- comparisons between vegetation under identical lighting -- and to make make quantitative comparisons in those scenarios. Without Ned's absolute calibration, we can't do other kinds of analysis, but it's suitable for many things as long as people understand its limits.

Yeah, I think the new camera tests have been a bit disorganized, with nobody taking lead on the idea of testing red filters. If we asked the Kits team to ship them to somebody else who's willing, would images such as the above ones I took (keeping the originals) be suitable for assessing? We could ask on the mailing list for a volunteer.


Yes, those photos are good. Ideal photos have some live plants, maybe some recently dead leaves, and some things that are not foliage. Outdoor photos are best.


Do you know if there is a place I can buy a usb camera already with the Wratten 25A, I got the original infragram which uses the blue channel, but I will be flying at altitudes where we will experience scatter. I think it would be fun to add this camera to a flight I have coming up.


Hi! The Mobius can be used as a webcam, and a converted lens is available in the store: https://publiclab.myshopify.com/collections/diy-infrared-photography


You must be logged in to comment.