Kush
What you are seeking is possible, but this may take several posts back-and-forth to work through the details. There are two major issues to address:
1) are you viewing the original pixel values, or have they been altered?
2) do you have an equation to convert accurately from pixel values to temperature?
Ignoring the radiometry (#2) for a moment, note that if you are working with a mosaic, in most cases the pixel values in that mosaic have been altered during the blending process to mosaic the separate images together. If the mosaicking software does not also apply a histogram stretch, this mosaic *might* work (you will have to test to verify), but my general recommendation, for best accuracy, will be:
- run the mosaicking process (Agisoft, Pix4D, Drone2Map, other)
- extract the photogrammetric metadata from the mosaicking software and use it to ingest the original images into ArcGIS using the Frame Camera raster type (see note below).
- This will allow you to orthorectify each image individually and not blend pixels together.
- You'd want to use the Nearest Neighbor resampling (not the default resampling = Bilinear)
- Add your equation to convert from pixel values to temperature as a raster function (if you're not familiar with raster functions, that's a separate discussion).
- add an optional color map if you want to simply identify temperature visually by color (the raster function w/o color map would allow you to click on a pixel to get temperature)
Now, NOTE that the online documentation for the Frame Camera raster type is not as detailed as I would like, and I'm actually working on a workflow specifically for it now. If you don't need it urgently in the next few days, I hope to get this completed and posted online soon. In addition, the step "extract the photogrammetric metadata from the mosaicking software" is not trivial, so I won't pretend we can do the above steps in just a few minutes.
My recommendation is to address the radiometric problem first. If you can't find an equation to give you good temperatures, then the workflow I mentioned above is not important.
I can't determine if the equation you show is valid, but I'd start with the FLIR documentation to determine if they provide one. Once you have an equation, you need to do some testing with individual images - nothing from the drone that has been rectified into a mosaic - these should be images taken on the ground, with controlled temperature objects that you can physically measure - e.g. maybe multiple coffee mugs with hot water in 2 of them, ice water in 2 more. If you cannot detect these temperatures with relative accuracy, attempting it from the air is a waste of time.
Two other comments:
- Read about emissivity. Depending on the surface material (e.g. concrete vs. metal), two objects at the same actual temperature will have different apparent temperatures in your thermal IR imagery. Nothing I've said above accounts for this, so you'll have to decide if your landcover material (grass, pavement, asphalt shingles, glass windows, water, etc.) must be mapped to account for emissivity before you try correlating to temperature. (Any equation you find for converting pixels to temperatures must either have an emissivity parameter, or you need to understand it is giving you only the apparent temperature).
- You can test this in the lab by using two different coffee mugs with hot water - get one metal mug, and for the other use a dull (not glossy) ceramic mug, and then put some masking tape or carpet tape on the lower half of both mugs. When you put in hot water, let the mugs sit still for ~3-4 minutes before recording the temperature and taking the thermal IR image. My prediction is that the masking tape areas will appear to be similar temperatures on both mugs, but the metal and ceramic surfaces will appear different temperatures.
- My comments at top about not using the orthomosaic created by Agisoft/Pix4D/Drone2Map may not be a concern. If you work through the challenge of finding an equation that works properly, the pixel values that you see in the blended orthomosaic might still be accurate enough for your work - you will have to determine that.
Cody B.
