How to generate a Thermal Orthomosaic / Map with radiometric information

How to generate a Thermal Orthomosaic / Map with radiometric information

The principals of photogrammetry are well documented and need not be explained in detail, however with the continual miniaturisation and improved resolution of thermal cameras begs the question: Can you apply the same principals of photogrammetry to thermal images? The short answer is yes, however the idea becomes more difficult when you attempt to retain the radiometric information. Below is a workflow that will allow you to generate thermal Orthomosaic and retain the radiometric information. 

Most radiometric sensors like the Tau 2 allow for 640×512 resolution .rjpeg images to be captured. The radiometric data however is not contained within the pixel values, rather in the meta data sitting behind them. As such, when thermal images are stitched together to produce an orthomosaic in photogrammetry software such as Pix4D, Metashape, or Context Capture, the radiometric data is not retained. The resulting orthomosaic will not be useful as a map of temperatures unless you know what temperature correspond to which pixel values (0 to 255).

The workflow is summarised below then explain in more detail.

  1. Import .rjep images into chosen photogrammetry software (Keep them in greyscale). 
  2. Run photo alignment, sparse point cloud generation, or aerotriangulation and the dense point cloud as you would for a normal aerial survey. 
  3. Use “runnable”script to batch change temperature range to something suitable.  Download here.
  4. Replace the images imported into your photogrammetry software with that with the edited and known temperature range. MAKE SURE THE FILE NAMES MATCH EXACTLY!
  5. Generate your orthomosaic using your chose photogrammetry software. 

This workflow produces and orthomosatic with know maximum and minimum values. As long as the colour gradient is linear (greyscale), then we know that temperature values fit evenly across gradient between our maximum and minimum values.   The resulting Orthomosaic can now be visualised using a stretch colour palette in software such as ESRI ArcMap. 

This process is not designed to make aesthetically pleasing images, rather more informative maps.

Processing the data in this way also results in maps being vulnerable to environmental factors during data collection such as clouds cover (Which can be seen in the attached image).