GCP error margin 20-30 cm

[Dukytony]

Hi,

I'm currently trying to understand a very unsual problem i'v had with metashape recently.

On a few projects both small (Poled camera) and huge scans (UAV),

Metashape indicates an error of around 20 to 30 cm in X and Y and sometimes Z

I'v taken the pictures as I always do without problem.

GCP coordinates were calculated first using a GNSS station. I thought at first that was the GNSS accuracy could be the problem so I took them again with a total station and same error is occuring.

Weirdest thing is that a few other scans done with the same method (GNSS) are fine...

I have litterarly no idea what i'm doing wrong. I'v been using metashape for years...

Please help me figure this out guys!

Thanks a lot!

[JMR]

some more about the circumstances can help to find possible causes.
Do cameras geotag images? likely in uav, but not so much in pole mounted camera...
if so... what is the camera centers accuracy in ref settings?
what is the  gcp accuracy settings?
what the overlap?
how is like the distribution of GCP with regard to the mapped area?
what about cameras pixel error after alignment?
what about tie points covariance?
Is the issue happening always with the same camera/lens? if so, have you checked the camera?
is that the same camera/lens that you have used for previous successful projects?
does it have rolling shutter? (for example, note that some Panasonic cameras, even if they have a mechanical shutter, they use electronic shutter for speeds beyond 1/500)

I hope we can help if you if you provide details

[Dukytony]

Do cameras geotag images? no it's not linked to geotag (poled camera without gps and i deleted the gps info of the uav images)
what is the  gcp accuracy settings? 0.005
what the overlap?
how is like the distribution of GCP with regard to the mapped area? I v put around 10 gcp around my area like i'v always done
what about cameras pixel error after alignment?

Points   293,807 of 314,345
Point colors   3 bands, uint8
Key points   No
RMS reprojection error   0.510024 (0.668583 pix)
Max reprojection error   1.54764 (19.6241 pix)
Mean key point size   1.36191 pix
Average tie point multiplicity   3.5013
Alignment parameters   
Accuracy   Highest
Generic preselection   Yes
Reference preselection   No
Key point limit   40,000
Tie point limit   4,000
Guided image matching   No
Adaptive camera model fitting   No
Matching time   1 hours 6 minutes
Matching memory usage   2.17 GB
Alignment time   11 minutes 20 seconds
Alignment memory usage   126.68 MB

what about tie points covariance?
Is the issue happening always with the same camera/lens? if so, have you checked the camera?

Its not the camera i use usually but the problem happened only on some projects I used it with. And also the same problem happened with some UAV images i got from a hired guy. So i don't think it's a camera problem.

is that the same camera/lens that you have used for previous successful projects?
does it have rolling shutter? (for example, note that some Panasonic cameras, even if they have a mechanical shutter, they use electronic shutter for speeds beyond 1/500)

The pole camera is a panasonic DMC FZ300
UAV was Phantom 4 pro.

I usually use a nikon 5100 never ever had that problem....

Thank you for your help!!!

[JMR]

Some comments to your details.
You should not use highest accuracy with your camera. High uses full image size and is actually the highest recommended setting for alignment (from the user manual "Highest accuracy setting is recommended only for very sharp image data and mostly for research purposes")
FZ300 is a good camera for many purposes, but you need to turn off the OIS otherwise it can ruin your photogrammetry project. Also autofocus should be disabled and focus set manually once according to the shooting distance for best results.

Estimating tie points covariance would give you chance for filtering and blunder detection. Also by analyzing the pixel errors in the right most column of your ref table, you would be able to see if photos are badly aligned. If error are clearly over 1 pix, then it is likely the alignment is not right.

I think you should not have cleared gps info from uav images. If you want to ignore global position, do it after alignement by making null the global transform. but during alignment, geotags will be helpful. I have never seen errors like 30cm in any P4Pro dataset.

 

[Dukytony]

So the UAV problem is solved. I just aligned with "high" and adaptive camera fitting model and it's all good now.

BUT

I still have problems with my lumix fz300. Same error margin of 20 to 30 cm with some scans. I disabled the OIS but that doesn't seem to be the problem.

All pictures are taken with the same focal so I don't understand what's going on...

[Dukytony]

here is what it looks like...

Please help!


thank you very much to all of you!