Very high error of markers

[Manuel]

Hello Guys,

first i want to thank the developers and their team for this great piece of software.

Currently I am using Photoscan to construct a 3D model of a little historic tower (4 x 4x 6 m).
There are 147 images taken and in addition 14 natural points (corners) surveyed with a total station, so their accuracy  is pretty high.
I placed the markers in all pictures, entered their coordinates (local) and ran the alignment with these. The RMS of the chunk is at a satisfying 0,5 px.

But when I look in the Reference tab, the markers have a very high error, mostly in X and Z axis (see screenshot)




- I cant explain myself that, i placed them very accurately in every picture and doublechecked that. So maybe someone of you can do that.


- Also, in the Reference Tab there are no coordinates and thus no errors of the cameras (images). Do I need to enter the coordinates for each camera maunually (i didn't survey these) or how do I get to them?


- In the Referene Settings, does the value for "Camera accuracy (m)" mean at which accuracy the positions of each camera has been surveyed?

- Is it recommended to load the markers after the alignment and optimisation?



Image: placed markers in one of the 147 images


Thank you in advance!


Manuel

[Alexey Pasumansky]

Hello Manuel,

I can suggest to check that the coordinate columns were not mixed on import and that the coordinates are not in geographic system, otherwise select it in the Reference pane settings.

After that you can try to reference the model using smaller number of markers starting from four) to see if there are any outliers that are causing large errors.

[Manuel]

Thank you for your answer.
The coordinate system is set to local, so this can't be the problem.

I checked and unchecked some of the markers and so came to the result that the markers at the top of the tower are problematic, expecially the z-value:


the value z-error is about twice as high as the height of the tower.


So when I switch to the model, the whole model is flipped upside down (see screenshot; it was taken with predefined front view)





As you can see the z-axis is pointing up, as the model is upside down, although the markers in the upper parts of the tower have higher z-values than the ones at the tower's bottom

[Alexey Pasumansky]

Hello Manuel,

What happens if you re-load the coordinates and swap X and Y columns?

[Manuel]

Thank you Alexey, that solved the problem