Support Website Contact Support Blog

Thermal .JPG Image Walkthrough

Is there a walkthrough on how to process thermal images that are only .jpgs and do not have radiometric data?

I am trying to create a 3D Model of a structure but want to use the color-mapped thermal images as the pointcloud colors/mesh texture.

I have collected thermal images (just .jpgs) using a Mavic 2 Enterprise Dual (M2ED) thermal camera (It is a FLIR Lepton 3.5 160x120). I have used many different settings but the initial processing of the images always fails due to failure to calibrate enough cameras.

Processing Settings Attempted

  • Auto-camera settings of FC2403_0.0_640x480 with Processing Template “Thermal Camera”
    – Initial Processing failed due to “No Calibrated Cameras”
  • Auto-camera settings of FC2403_0.0_640x480 with Processing Template “3D Models”
    – Calibrated ~4% of cameras
  • Created a camera model and set pixel size and focal length based off of FLIR Lepton specs with processing template “3D Models”
    – Initial processing calibrated roughly ~2% of cameras
  • Camera model based off of FLIR Lepton, and processing template “Thermal Camera”
    – Initial processing failed due to “No Calibrated Cameras”

Questions

  • Is there a walkthrough for doing thermal processing with only .jpg color-mapped images?
  • My thermal image color scale is gray -> yellow -> orange -> red -> white, would I need to take this into account in my settings? Would I be changing this in the “camera model bands” options within camera model?
  • I also have RGB images that I have used a to make a 3D model that I am hoping to merge with my thermal images project, as suggested here, but since my thermal images don’t have radiometric data, should I try and process them all together?
  • Is it possible to used images from the M2ED or do I just need a different camera?

Info and links I looked at when doing the processing attempts listed above

Thank you for your help,
DJ

1 Like

Hi,

Thank you for your detailed explanation, makes our life easier!

The workflow for normal jpg images for thermal (RGB), tif and rjpeg are the same. It is just that for rjpeg, we have a FLIR SDK that converts the images internally in Pix4DMapper to tif and uses the radiometric information. For tif and rjpeg, the final map you get has temperature values in the pixels. For RGB, you will get a map which can be used visually for pointing out changes in temperature, however, it will be a RGB map (3 bands).

For your settings:

The camera model consists of the model name_focal length_resolution. You can see that you have 0.0 as your focal length which is incorrect.It means somehow the lens is not picking up the focal length value. Pix4D reads this from the EXIF. You would need to contact FLIR to see why this is happening.

For the processing, make sure the parameters are correct. Use thermal template and make the following changes:

  1. Use standard calibration
  2. Enable geometrically verified matching

Do you also have a good overlap? That is the most important as thermal images have low resolution. What was the flight height?

For the color scale, you will need to change it in the index calculator if you want to see the changes in the reflectance map (use the grayscale index which will give 1 band using weights from red, green and blue): https://support.pix4d.com/hc/en-us/articles/202560519-How-to-edit-the-Color-Mapping-in-the-Index-Calculator

For 3D mesh, you can make the changes in the raycloud editor: https://support.pix4d.com/hc/en-us/articles/360039283971-How-to-capture-and-process-data-with-Parrot-ANAFI-Thermal##How%20to%20visualize%20the%20thermal%20mesh

You will need to process all the RGB images together in one project.

Let us know how it goes.

Momtanu,

Thanks for your reply. My overlap is at ~30-40% and my images are taken at heights ranging from 26-40 meters. Your suggestions of using standard calibration and geometrically verified matching improved my results but they were still unusable. I will re-take images with higher overlap, shooting for 90% and then reprocess and see how they come out.

Thanks,
DJ

DJ,

The overlap seems too low. I would suggest keeping the flight height at minimum 40 m and then lower the height if you get good results at this height.