Dense cloud wall surface covered by dupliacted, poor cloud

[Damian]

Hello

I have an issue with dense cloud of a building:

We have flew over renovated XIX-century factory to create a detiled Mesh-model. Flight was planned to obtain just vertical and diagonal photos (no typical horizontal photos with roll and pitch = 0). Just walls, not much of a roofs. During the processing we have discovered, that the general view of a building varies. Some walls look excellent, while other are just poor [Picture 2].

What is more, the poor part of dense cloud turned out to be mostly a duplicated surface [Picture 3, 3a, 3b] , that covered the good one (as you can see in Pic 3a-3b). The whole building looks very bad - but if I cleared the duplicated surface (see the light bricks on Pic 3), it turns out to be perfectly scanned.

While all the walls look similar and there were no difference in flights, what might be a reason and how to get rid of this poor cover? It really affects the mesh texture and tiled model.

Looking forward to any suggestion! Thanks!

[Alexey Pasumansky]

Hello Damian,

Can you re-upload missing pictures (pic.3)? Also please post a screenshot of Align Photos task parameters from Chunk Info.

Picture 2 looks like there are some sub-sets of misaligned images or for some cameras the depth data has been corrupted. The latter should be visible via Show Depth mode, if you review the cameras for the related facade.

[Damian]

Here are the missing pics. This duplicated surface is what I am worried about, as gros error points can be easily deleted by me.

[Alexey Pasumansky]

Hello Damian,

I can suggest to make a test and re-align the same dataset with Reference preselection disabled, providing that all cameras are unchecked in the Reference pane. Also I would suggest to disable Adaptive camera model fitting.

After initial alignment is finished, you can check on cameras in the Reference pane again and perform optimization (check on all calibration coefficients, but disable fit additional correction and adaptive model fitting options).

It would likely take longer time to align, than your initial try.

[Damian]

Hi


I proceeded the re-align and some improvement in dense cloud can be noticed. However, it is sitll not perfect and still not easy for manual clean. Could you look into it again and give me some advice?


In attachement you may find Camera Calibration settings and reference pane. I also attach captured views to show the remaining surface.

[Paulo]

Hello Damian,

if I remember correctly, the L1Dc camera (Mavic2Pro) has a rolling shutter, so it would be good to see the adjusted tab from your camera calibration to see how the values are. What is your exposure time and is drone moving during shots?... Maybe rolling shutter could help...

The other thing is that if your 4 markers are accurate and have no blunders then in no way you should have a 2 m error after optimización.

Something is fishy... here

[Damian]

I would not pay attention to markers, as they are not mesured by GNSS, they were used just to get some georeference. I should have uncheked them now, for visualisaion.

[Dieter]

This result cannot be if the recordings have enough overlap.

 I also work with the Mavic and I have never had such bad results.

 Offer the pictures for download, then I can help you.

Dieter

[Damian]

Dieter,

here you may find data:

http://gofile.me/4IceV/m3fk4BW4k

As previously mentioned - I did not optimize realignment with markers - they sould have been unchecked when printscreened.

Loking forward to getting any advice from you! Thanks!

[Dieter]

I'm downloading it, it's taking time...

What I can already say: 1/30 shutter speed is not enough for interval recording. Half of the images are blurred. I fly without an interval and have a shutter speed of at least 1/100, preferably 1/200 if possible. I set ISO to AUTO.

I (and Metashape too) prefer noisy images to shaky ones.

Dieter

[Dieter]

You can calculate the quality of the images in Metashape, from 80 is a good value that I prefer to work with, your images are 50% below this value.

In order to achieve high-resolution ortho images or a highly accurate point cloud, you have to work on your recording technique with your drone. That just by the way.

I calculated the images, you can see my settings in the attached image. I'm working with the 2.02 beta, because there were problems with scan files in earlier versions, but that shouldn't affect this calculation here.

The result is good in terms of quality, despite the bad pictures. I can't judge why you have such bad results, just try a recalculation with my settings.

These settings are my default for any kind of alignment in Metashape.

The calculation takes a little longer, but the results are better, especially in poor recording situations with few identical points on the photos.