# Question: First Drone Flight and White Balance Issues

amck asked on September 14, 2015 22:42
1,881 | 0 answers | #12215

I recently purchased a Filter Kit and installed the included #2007 blue filter on a new ELPH 130IS. Yesterday I attached the camera to my quadcopter and collected ~600 images covering 60 acres of mixed agricultural land. Lacking anything other than a random solid blue Mead folder that was lying around, I set the white balance on the camera with the frame filled with the folder. The images mostly came out great and processed flawlessly in Photoscan, but there seemed to be an awful lot of light coming into the Green channel. In fact, the green histogram is nearly identical to the red channel of every image. I assume this is an issue with how I set the white balance.

Does anyone have any specific recommendations with what medium is currently preferred to correctly minimize the green channel of the starting image? I've seen suggestions of test paper and led lights, but I'm not sure what the best option is for most field settings.

I'll post some of the full NIR 3D models when I get everything processed.

I see what you mean about the green channel being almost as bright as the red channel. The Rosco 2007 filter blocks almost as much green as red, so both of those channels are going to be dominated by NIR. A different white balance setting could darken the green channel some, but it is still going to include mostly information about NIR. The green channel does not have to enter into computations of NDVI, so it might not be so important.

A more important issue might be that the blue channel is very dark. That provides very good separation between the VIS (blue) and NIR (red) channels so NDVI values will be plenty high. But the blue channel is so dark that most plant pixels and many other things have a similar brightness. This might be causing poor distinction between plants and non-plants in the NDVI values.

Many non-plant things (like the round bales and deck) have higher NDVI values than they should. Some of this is due to the nature of the blue filter which uses blue light for VIS and therefore does not allow healthy foliage to be distinguished from some dead things (including wood and dead foliage). Some of the problem might also be due to the particular white balance setting -- the red and blue channels probably don't have to be so well separated.

You could try another white balance with another blue paper (in the shade under a blue sky) and see how it works. Or you could try a red filter which will generally provide better distinction between healthy foliage and things like wood and dead leaves. There are some advantages to using a blue filter, but only if the cutoffs are very steep on either side of red. The Rosco 2007 does not have very steep cutoffs.

Chris

Thanks for the reply Chris. Your observations are very insightful. My thoughts are currently toward reproducing some of your tests with hooking up my camera to the public labs spectrometer and measuring a variety of different filters. My main goal is to use this camera to begin mapping fields to look for archaeological features that may be visible by subtly affecting plant growth. The more I'm reading, I suspect that a red filter will give me a deeper NIR signal that may be better than the ~740 nm range that I think I'm collecting now.

I'm not sure how different the NIR signal will be when using a Rosco #2007 vs. #19 or other red filter. Both will allow capture of some distribution of wavelengths between 700 and 900, but the exact distribution will depend largely on the sensor and Bayer filter in the camera. The primary difference among those filters is how much VIS contamination will be in the channel used for NIR. Red filters that don't pass much blue or green will have rather clean NIR signals in the blue channel. The Wratten 25 should give such a result, and Rosco #19 might be similar. Rosco blue filters will have mostly NIR in the red channel but will always have some red (and probably blue and green) mixed in with the NIR and it's really hard to know how much. Neither blue nor red filters will have a very clean VIS channel, since NIR probably contaminates them all.

So without going to a pure NIR filter (e.g., Wratten 87) the purest NIR channel will probably come from a good red filter like Wratten 25. That will also provide a red VIS channel (with considerable NIR contamination) and a green channel with mostly NIR in it. If you think another VIS channel might be helpful distinguishing vegetation patterns, a yellow filter like Wratten 15 will produce a green channel with some green in it. The other channels will be like the Wratten 25 except that the blue channel will not be as pure NIR.

Which of these filters will produce the best discrimination among slightly different vegetation types is hard to predict. The time of year and time of day are probably equally important if not more so. On the right day, a normal RGB photo from the right angle might reveal as much as any NIR photo. Part of the advantage of an unmodified camera is that it's optics have not been compromised and more light is getting to the sensor so ISO, shutter speed, and aperture can be set for better clarity.

Which will come first, a $1000 drone that can carry two good cameras (RGB and NIR), or two$500 drones that carry single cameras and fly in perfect formation three feet apart (and sync camera shutters)?

Is this a question? Click here to post it to the Questions page.