Forum

Author Topic: FLIR Duo R - Thermal - Processing as a Rigid Camera Rig - Multi-Camera System  (Read 1593 times)

jmgc

  • Jr. Member
  • **
  • Posts: 62
    • View Profile
Hello Metashape Fellows,

I'm bringing again to the discussion the processing of images captured with the FLIR Duo R.
There have been a number of reports from other users who were able to obtain thermal orthomosaics, and from others who were not so lucky. However, as far as I know, there is no generalized procedure for doing this.
Presently is possible to obtain a separate RGB and Thermal ortomosaics. The main problem is to get both co-registered.

** Some time ago I've shared here a procedure where RGB images (cut to size 1440x1080) were processed first, and then replaced in the project with the thermal images (resized to 1440x1080). It was not perfect, it had clear misalignments, but it already gave an idea of ​​the thermal panorama of the covered area. Presently I don't recommend this option..

** A more consistent procedure is to merge the RGB images with the thermal images (both 1440x1080), creating a single multi-spectral RGBT image. This article explains and demonstrates this procedure: https://www.mdpi.com/2072-4292/11/11/1365.
I made several attempts in this way, the procedure surely becomes more seamless but still misses co-registration between bands. The RGB orthomosaic appears correctly but the Thermal shows evident stretching misalignments. So, improvements must be done.

** Currently, as I've mentioned, processing only thermal images produces good results. This is already a big step forward.

** Lately, using the multi-camera system processing approach, my expectation would be that this process could work with the FLIR Duo R. Unfortunately, the results are still not satisfactory: the thermal orthomosaic always presents stretching misalignments.
In comparison, I did several processing with sample images captured with Micasense Altum (https://micasense.com/sample-data-altum) - 5 bands + Thermal - always with good results. To try to match the processing conditions to the FLIR Duo R, I removed all the orientation information and characteristics of the sensors that are registered in the EXIF ​​of the Altum images. Even so, the results were good. The thermal orthomosaic stitching is good and and the co-registration between the multi-spectral bands and the thermal band (orthomosaics) loses very little quality.

I'm attaching some results from my processing with the FLIR Duo R and I'm providing the image dataset in use, asking for collaboration of the interested users and Metashape moderators, for a proper solution.

Image dataset RGB + Thermal: https://meocloud.pt/link/453b2b19-5475-443f-9106-96745c93b105/FLIR-Duo-R/
« Last Edit: February 08, 2021, 04:45:02 PM by jmgc »
José Miguel Campos
Geospatial Specialist
UAS Operations Manager

jmgc

  • Jr. Member
  • **
  • Posts: 62
    • View Profile
Any thoughts?
Anyone to process this imagery and share the results?
Thanks.
José Miguel Campos
Geospatial Specialist
UAS Operations Manager

JoseASanchez

  • Newbie
  • *
  • Posts: 5
    • View Profile
Hi José Miguel,

I'm trying to do the same with Zenmuse XT2. Send you a private message

tuncaemre

  • Newbie
  • *
  • Posts: 1
    • View Profile
There is another possible way to generate orthomosaic is using black and white stones for gcp. Before flying, get wetted to white ones. In the thermal image you can easily recognize these points. This is what we are using to obtain orthomosaic.

JyunPingJhan

  • Newbie
  • *
  • Posts: 1
    • View Profile
Our approach is to co-register original images that is to correct lens distortion and misregistration errors of RGB and TIR images in the same time.

Therefore, we cna use RGB images for camera alignment, generate high resolution DEM, and rectify images. Then, we can use same EO of RGB and DEM for rectifying TIR images and obtaining spatial co-registered multi channel RGB/TIR ortho images.

You can find some samples here.

https://drive.google.com/drive/folders/1dBnsBU9x4-hHyz47kQfhGnRmz4XWJnIi?usp=sharing


 

adrien_gbx

  • Newbie
  • *
  • Posts: 7
    • View Profile
Could you please provide some insights on how you did this?

Is it something published somewhere else?