Confused on NDVI Workflow for 2 Canon S100 cameras

[TXPE]

I have 2 Canon S100 cameras.  One has the filter installed.  Both cameras are mounted to UAV but don't always trigger at same time/location.

I'm trying to figure out the proper workflow to generate the NDVI map.  Because it is not a true multispectral camera, I'm not sure how to create the formula to pull data from both mosaics (RGB & NGB).  Actually, I'm not even sure I'm setting up the project correctly.  I have one chunk for RGB and another chunk for NGB.  I hope this is correct.

Can anyone share the correct process to create index map using two different images?

[zapotek]

Did you get this figured out? I have a similar question.

Thanks!

[Alexey Pasumansky]

Hello,

Multichannel chunk approach implemented in the actual version of PhotoScan assumes that the sensors are shooting simultaneously and are mounted close to each other. Having two cameras that take photos in the different moments of time for the same area is almost the same, as two separate flights made with the different cameras, so the multichannel approach is not applicable to such type of input data. It means that the images related to each camera should be processed and georeferenced independently in separate chunks, exported and then merged into multichannel raster in the external application. To minimize the possible shift between the channels it's recommended to use precise GCPs for chunks' georeferencing.

[TXPE]

I created two different raster images using Photoscan, one RGB and one NGB.  I then load those two into Global Mapper.  In GM, I can create a raster calculation that pulls the R channel from the RGB and the N channel from the NGB to create the NDVI map.

[GrinGEO]

Im still too interested in this kind of workflow. Im wondering what type of camera do you need`?

is it enough to have one RGB and one infrared camera?
or do you need more cameras with different filters for infrared & near infrared?

Can you please explain the processing of how you interpret the data from only RGB + Infrared pictures?

Would be vey thankful

[stihl]

GrinGEO,

Perhaps it would be wise to follow a course or study regarding "Remote sensing" and perhaps a Photogrammetry lesson as well.
Basically you're asking very specific information which is hard to expain without using jargon.

Modified cameras can be bought which are altered to store information in the NIR band instead of the Red band.
What you'd want to do is mount two cameras, a regular RGB and a modified NIR model, and capture the terrain in question. Then generate a seperate RGB Orthomosaic and a seperate NIR Orthomosaic in Photoscan. Photoscan will know how to deal with NIR images.

TXPE has answered that he uses GlobalMapper to extract certain data from one data set  and incorporate it into another data set. The NDVI formula goes as following; (NIR - Red) / (NIR + Red).

So what you do in Global Mapper is load both the RGB and the NIR orthomosaic files and tell the software to replace the R in RGB with NIR from the NIR Orthomosaic.

Dataset 1: R¹ G¹ B¹
                                              ---> ( NIR² - R¹ ) / ( NIR² + R¹ )
Dataset 2: NIR² G² B²

Then you let GM calculate a new raster index based on this information with values ranging from +1.0 to -1.0.

[GrinGEO]

unfortunately we dont have any lessons here in our region... Im fine with the photogrammetry stuff and are getting good georef results in Agisoft. Unfortunately I can't find any informations regarding NDIV calculations, unless the infos you get on MAPIR with their cameras....

If you can give me any infos... Im happy with it.

Im still wondering how a NDIV can be generated only from RGB+IR instead of IR+NIR camera.

[stihl]

GrinGEO,

Make sure not to confuse yourself with IR and NIR. Infrared (IR) is not what's used by these modified cameras. These cameras capture in the Near Infrared(NIR) band. This is different from IR.

A NDVI map can be generated when you have two Orthomosaics of the same terrain. One Orthomosaic has to be a regular R-G-B photo and the other has to be a NIR-G-B. With these two datasets you can generate a new Orthomosaic that uses the NDVI formula.

[GrinGEO]

Thanks Stihl. Im not confusing IR with NIR and I understand the difference. Im just wondering why the MAPIR Cameras are able to record IR+NIR at the same time with a special dual-band filter splitting the signal into Red & Blue RGB channel.

And from my understanding, a NDIV image needs to be calculatet from (IR-NIR) / (IR+NIR)

so I was wondering if NDIV can be also calculatet from NIR + RGB (logically 2 separate Orthomosaics) only. If yes, Im more than happy since ill get my IR camera this days.

update
Maybe I misunderstood something due to a wrong translation. it seems I need the red channel & NIR... than I should be fine with 2 cameras (one RGB & one NIR)

[stihl]

GrinGEO,

There is no such thing as NDIV, only NDVI. The formula is described in my post above.

Give this article a read, it should clarify some things for you:

http://agribotix.com/blog/2014/06/10/misconceptions-about-uav-collected-ndvi-imagery-and-the-agribotix-experience-in-ground-truthing-these-images-for-agriculture/

You are correct that you only need a NIR or IR,  and a regular RGB Orthomosaic to create a NDVI index map.

[GrinGEO]

I meant NDVI, just typed wrongly on my keyboard 

[tgilbert]

I don't think that's quite correct. I use both RGB and NIR versions of the Canon S100 and I only need the NIR version to calculate NDVI using the raster calculator. My NIR replaced the Red channel, so I use the standard raster calculation for BGNIR. I've not needed to use my RGB imagery to create NDVI.

If there is a way to get better NDVI by combining the RGB images, I'd be interested to hear about it. But it doesn't seem to be critical as the results are comparable to output from SMS Leader or Drone Deploy NDVI which also use only the NIR.

[stihl]

I'm afraid you don't fully realize how a NDVI is generated.

An NDVI by itself isn't an image. It's an index and its visualization can be radically different depending on the color pallet chosen by it's user.

If you have a NIR-G-B image (RGB with the R channel replaced so the whole image has a pink hue), you will need the R channel from another RGB photo to properly use the NDVI formula.

Since you say you don't need it, I'd love to hear where you're getting your red channel from since there's no red in your NIR image.

What DroneDeploy is likely to be doing it generating a so called "False-NDVI", have a thorough read of this. If you're interested you should look up why you need a separate NIR and RGB to properly calculate a full range NDVI.

[anyalamb]

Hi stihl,

Just want to clarify -- DroneDeploy supports both true and false NDVI.

If you captured your imagery using a camera that includes NIR, when you look at your map in DroneDeploy you specify the band order of your camera (e.g., NIR-G-B) and then choose from a number of indices that you can apply (including NDVI).

However, if your map is made from RGB imagery and you try to apply the NDVI index, it will be false rather than true NDVI. One thing to note is that DroneDeploy also includes the VARI index, which is specifically intended to highlight plant health variability using RGB imagery.

Here's a support page about how to select the correct band order of your camera and apply an appropriate vegetation index in DroneDeploy: http://support.dronedeploy.com/docs/ndvi-band-order

Thanks!
Anya

[Alexey Pasumansky]

Raster Calculator in PhotoScan works in a similar way, as it allows to input the formula for the index calculation that may include any of the original image bands.