Rolling shutter correction for Anafi

[bernarde]

Hi all,
I tried the rolling shutter correction parameter with an Anafi but the resulting orthophoto shows artifacts.
Having a look at the RS repository, I realised that data for the Anafi is not present even if this message seems to provide it :

Phantom 4 Photo Request [ Rolling Shutter ]

OK, I got a flashing light set up courtesy of the engineers here at the Norwegian Polar Institute. I used a blink frequency of 2kHz (0.00025s on, 0.00025s off), and shutter speed of 1/4000 For both ANAFI USA and ANAFI thermal I count the same number of bands in an image - 80 pairs / 160 total. These aicraft both use the same 21MP sensor, so this makes sense. A sample image (ANAFI USA) is attached. The next step I’m not sure I understand well enough. If I take 1/ ( 160/4000 ) I get a reasonabl…

So I tried with 160ms as readout time but there are still the same kind of artifacts.

I will try to make my own measurements but as anybody successfully applied the rolling shutter correction to some anafi datas ?

Thank you all for your feedback

[amwphotos]

The equation posted was this:

readout = (1 / F * 1000) * L (ms)

F = frequency (hz)
L = lines count

so…

(1/2000*1000) * 160 = 80ms readout time

Try that and see if it improves things. If you look at the quality report for both settings you should be able to see what difference has been made

[bernarde]

I was looking for the formula thank you.

Alas it seems globally better but there are still weird artifacts.
I will have a deeper look on Monday and will provide you some screenshots or the reports.

Regarding the report what part/data exactly should I be looking at?

[bernarde]

OK, here are two reports :

• one with rolling shutter correction with a 80ms readout time :
  ODM_report_20230227172836581.pdf (7.4 MB)

• the other “normal” processing:
  ODM_report_20230227175219783.pdf (7.6 MB)
  

The same artefacts appear with 160ms readout time.

Any idea or what I should be looking at ?

[amwphotos]

The Z axis looked like it went a bit crazy with the rolling shutter correction, but not doing any correction seems to have given a respectable result. I’d just not bother with the shutter correction as it’s obviously causing issues in this instance.

[bernarde]

Thank you for your answer but I would be curious to find what causes this strange behavior.

We are running ODM automatically inside a pipeline here (automatic flight, data transfer, reconstruction, filtering, reprojection and importation). That is why we can not “afford” to look at each dataset and it’s output to decided weather or not we should rerun-it with or without the rolling shutter correction.
If I can’t find a reason I will be forced to “not enable” it for all reconstructions to prevent this case.
Would be a shame I think…

[adamsteer]

Hiya - I did the flicker test with ANAFI Thermal and ANAFI USA and got 160 ms. My results haven’t been so great applying a shutter correction, so I never updated the database. More work on why is funding-dependent

[smathermather]

Strong suspicion: with longer shutter times and certain types of movements, rolling shutter corrections using this approach break down. 160ms is pretty long. I would recommend keeping it off for Anafi.

[adamsteer]

That’s my gut feeling also. I was flying sideways quite a bit…

[smathermather]

@adamsteer

I was flying sideways quite a bit…

[bernarde]

Ok, than you for this advice.