When I try to calibrate ‘Evo II Pro (Autel Robotics)’ images, failed.
- use PIX4Dmatic (ver 1.18.0)
- 64 images.
- success when use PIX4Dmapper
I think internal camera information is not include on PIX4Dmatic.
Would you like to check ?
Hello @seiichi_takayama, thanks for reporting the issue. It is hard to know what went wrong without having the data. Would it be possible to share the dataset with us? If you prefer to not post it on the community, let me know here and I’ll reach out to you by email.
Thank you your reply.
Today’s observation’s images are failed, too.
This is images.(approx. 1.4GB , I will delete after you downloaded or within 5 days.)
horizontal : JGD2011-CS09 (epsg 6677)
vertical : JGD2011(vertical)height (epsg 6695)
@seiichi_takayama thanks for sharing, indeed, the calibration does not seem to work in PIX4Dmatic. At a first glance, the content of the images seems difficult to reconstruct: flat area with grass and the road seems to be slightly overexposed, maybe there is even some blurriness.
Would it be possible to share the quality report of the PIX4Dmapper project for which you mention that the calibration worked on these images? This will help me understand what settings you have used. Thank you!
Thank you your reply.
This is the report processed by PIX4Dmapper.
project_01_H42OL80SL70_report.pdf (1.1 MB)
Thanks for sharing! After some further investigations, I believe that the calibration pipeline that is currently in PIX4Dmatic is not meant for flat agricultural type terrains. These are tricky areas (flat, difficult to match keypoints,…). We will plan to experiment with more settings and maybe a different calibration pipeline to handle such cases in the future. One difference we noticed is that there is no automatic “rematch” in PIX4Dmatic’s pipeline. Sorry for the inconvenience!
Thank you your investigation and reply. I see.
I expect to next ( or next next ) version.
It is just the same for me, I got crazy before of reading this your post
We have tried to calibrate the dataset with the new calibration pipelines and templates we have added in version 1.25.0. While doing these tests, we realized that the problem is not the calibration itself, but that it is a problem in the EXIF of these images. It’s an image with a perspective sensor model specified in the exif since it has the tag
1Xmp.Camera.ModelType XmpText 11 perspective
however, then it has null focal length tagged:
1Xmp.Camera.PrincipalPoint XmpText 17 6.566400,4.377600 2Xmp.Camera.PerspectiveFocalLength XmpText 8 0.000000 3Xmp.Camera.PerspectiveDistortion XmpText 74 0.000000000000,0.000000000000,0.000000000000,0.000000000000,0.000000000000
So the camera-creation behavior is as expected.
The project can get calibrated in matic 1.25.0 with default ISC template if I modify the exif tags
1Xmp.Camera.PerspectiveFocalLength XmpText 8 9.946491 2Xmp.Camera.PerspectiveDistortion XmpText 51 0.0665994,-0.211744,0.246277,0.00331114,0.000732718
We need to understand why there is a null focal length tagged. Is it an issue with all projects from this camera/drone? Is it from an outdated firmware version? For instance, can we find another project from this camera and check if it also has the same issue? Is this a known issue? Answering these questions may help to decide what to do because obviously, we could add some logic also to camera-creation to cope with this. However, before doing this kind of hacks we should understand if they are really motivated. The simplest would be that the drone manufacturer tags the right values in the EXIF of the images.
Thank you your reply.
My dataset is taken form drone which is latest version.
I think this problem is some kind of fault for this drone.
PIX4Dmapper fits this drone(Evo II) , but PIX4Dmatic doesn’t fits this drone(Evo II) , I understand.
Will you in the future modify the programming of Pix4Dmatic to allow the Evo2 Pro V2 drone to be supported?
Would you have a dataset to share with us from this camera? We could look into calibrating the camera and adding it to our camera database, this might solve the issue.
A good dataset with the following properties is required:
- Enough images (100 - 300 images).
- Non planar scene (not all objects having the same height)
- High overlap and rich texture.
- Several grid flights (at different heights) OR oblique imagery with sufficient overlap.
- If possible, with GCPs.
@teknik.drone your last replies seem to have an issue? I can see the following:
Is it possible that the camera (XT 705_5472x2648) of the EVO 2 PRO V2 drone from Autel Robotice, which is calibrated and referenced in Pix4Dmapper, is transposed and recognized in Pix4Dmatic, knowing that the EVO 2 PRO is a drone that has some commercial success?
Hi @teknik.drone, can you please provide me with a few images of that drone? We’ll look into taking the same model that is in the camera database of PIX4Dmapper to add it to the PIX4Dmatic camera database. The images are used to ensure the compatibility is correct, e.g. to test the import, etc…
If you provide us with a dataset that we can use to redo the full calibration, we can look into that. This would ensure we have a new camera model created and we don’t simply take the ones that were available in PIX4Dmapper. Here would be the recommendations (at the top of the article): https://support.pix4d.com/hc/en-us/articles/206065716-How-to-calibrate-a-Perspective-Lens-Camera