Radiometric (thermal) Orthomosaics - Photoscan vs Pix4D

[ascornelio]

Hello,

I'm interested in producing radiometric orthomosaics and have been experimenting with a public data set (https://www.rmus.com/pages/data-set-examples).

 I have run the data set with Pix4d (using its thermal template) and PhotoScan. Unfortunately Pix4d beats PhotoScan...
Results are attached.

This dataset consists of radiometric geotagged tifs, taken from the Zenmuse XT 640 x 512 resolution, 30 hz, 9mm lens thermal camera.

I'm wondering whether any one has any advice on how to produce better orthos using PhotoScan.
My steps for PhotoScan are pretty standard:
Align Photos
Build Dense cloud using Spare cloud
Build mesh using dense cloud and height field
Build ortho using mesh
Export ortho

The 9mm lens obviously causes a fisheye distortion on the images - i'm thinking about correcting this and then re-trying photoscan.
For some reason when I upload my images, then go Tools>Camera Calibration and change the camera type to Fisheye - no images ever align...

Any other advice would be greatly appreciated!!

Cheers

[selimerguden]

Hey ascornelio,

What settings are you using in Agisoft? I would recommend using High Settings for the Alignment  step and Medium Settings for everything else. I would also create a DSM and generate the orthomosaic from a DSM instead of a Mesh.

[Alexey Pasumansky]

Hello ascornelio,

That's what I've got as a quick test in the latest PhotoScan Pro version:

[Alexey Pasumansky]

The link to the project in PSX format is below:
https://www.dropbox.com/s/26mcls063x83gl3/PS_processing.zip?dl=0

The general steps that I've followed:
- after adding photos unchecked all cameras in the Reference pane,
- in the Camera Calibration dialog set the focal length in pixels to 100 (just a guess) and set to zero all other values in the Initial values tab,
- run Align Photos (High accuracy, Reference only preselection, 40 000 / 10 000 for key point and tie point limits), adaptive camera model fitting - disabled,
- only about a 1/4 of images were aligned,
- then I checked on all the cameras in the Reference pane, selected not aligned cameras and run Align Selected Cameras operation.

[Pkroks]

Sorry to resurrect an old thread, did anyone ever succeed at making a radiometric orthomosaic?

[JMR]

Yes we do. but with our PPK system (no ground control points). Also we usually use DEM out of an rgb camera images better than trying to build a good dense cloud with the thermal images.

[Pkroks]

Yes we do. but with our PPK system (no ground control points). Also we usually use DEM out of an rgb camera images better than trying to build a good dense cloud with the thermal images.

Can you help to explain your workflow for processing the thermal orthomosaic please.

Many thanks

[Aender84]

Hello, i'm also very interested in the Radiometric Ortomosaic. I managed to stitch the images together in a big tiff file, but that one is not radiomatric. is there any way i can output a big radiomatric image from Photoscan?

[DriesS]

I've been able to stitch a lot of of tiffs ('black ones') together with the workflow from this link:

https://juliageographer.files.wordpress.com/2017/05/agisoft-photoscan-workflow-for-flir-vue-pro-for-blog.pdf

without GCP's or RTK i still manged to get these images