High Error between Initial and Computed Camera Positions of Wall Scan

We have been attempting to do facade mapping using Pix4Dcapture and processing in Pix4Dmapper. The cameras are being incorrectly oriented and this is leading to issues in processing down the line. Would this issue be fixed with GCPs?

Thanks in advance.

1 Like

Hi @bsershen,

Welcome to the Pix4D community.

Please try to change the Internal Parameters Optimization to All Prior:

If the difference between the initial and optimized camera parameters is higher than 5%, the All Prior can be used to keep the computed values close to the initial values. This typically happens in datasets of flat and homogenous areas that do not provide enough visual information for optimal camera calibration.

Please let us know how it goes.

Hi Daniele,

That‚Äôs my go-to fix for when I have this issue for standard top-down mapping missions but this did not fix the issue. Left is the processing result with all default calibration settings and right is with ‚ÄėAll-Prior‚Äô. Both have geometrically verified matching checked.

In the quality reports, using the ‚ÄėAll Prior‚Äô method did drop the relative difference between initial and optimized internal camera parameters from ~5% to ~0% but it did not seem to lead to any improvement to the camera positions. The scene is still oriented at a weird angle.

GCIwallTest_report.pdf (297.3 KB)

GCIwallAllPrior_report.pdf (286.4 KB)

1 Like

Hi,

Thank you for the screenshots and the quality reports.
It seems that for your pictures, PIX4Dmapper is finding a better solution assuming tilted cameras. I have not seen the images but I imagine this could be caused by a repetition pattern on the wall.

Please try to add some tie points, in order to ‚Äútell‚ÄĚ the software which are the points that are referring to the same feature. Make sure to mark the points in images coming from the different ‚Äúflight lines‚ÄĚ.

Daniele,

The wall definitely is most a repeating pattern (see attached image). I added 7 MTPs to different surfaces on the wall, marking 16-30 images for each, but saw no visual improve after reoptimizing. When you say Pix4Dmapper does better assuming tilted cameras, does this mean I could get a better result using slightly tilted cameras? These were shot as near 0¬į as I could get. I‚Äôll attach the regenerated quality report.

Thank you for your continued assistance.

GCIwallAllPrior_report.pdf (276.0 KB)

Hi @bsershen,

Thank you for the reply and for sharing the image.
I see now that my previous message could have been misinterpreted.
I meant that PIX4Dmapper had found a better solution (from their point of view) than what is the reality.
This regular repeating pattern is really challenging. When you created MTPs, have you made sure that you marked them in images coming from different lines?

Also, try to change the Shutter Model to Global Shutter and process again from the beginning.

Let me know if this brings any change to your results.

Cheers,
Daniele

Hi Daniele,

Yes the wall is very difficult to map. I am processing another flight of the wall to see if the orthoplane DSM measurements are reproducible and fit for use in fa√ßade inspection, specifically checking wall depth variation (see attached photo). If so, I guess the orientation offset is a non-issue. I processed this 2nd flight with global shutter as well as ‚Äėall prior‚Äô and the step 1 results look good.

Thanks again

hi @bsershen,

Thank you for the additional information.

If you really want to use the incorrectly oriented model, make sure that there are no issues with the scale of the model.

If you are going to fly again this project, please also try to have flight lines all at the same distance from the facade.

If necessary, you can even try to re-orient your project:

Cheers,
Daniele

Hi Daniele,

Thank you for the tips. I was able to reproduce the variations in the wall with the 2nd flight but the scale was off (I also processed in ft instead of meters so that could also affect scale). I I tried applying scale and orientation constraints and reoptimized. It does appear the points were rescaled as the annotations no longer line up with the points, but the orientation is still not correct.

Thank you for trying again, @bsershen

If you are going to fly again, please also try to have flight lines all at the same distance from the facade and let me know how it goes.

Otherwise, try to create a new project and import only the images marked in the circle:

image