High issues in point cloud - two layers of points?

[Saltcity]

Hi everybody,

first of all: I’m really impressed of Pix4D. But in an actual project i have some problems.

First one has something to do with the output of high-values in the point cloud. We have flown with a Phantom 4RTK over an Area of about 315.500 m^2 and have taken nearly 5000 photos in 60 and 80 m above ground (cross).

The inner accuracy was very good. The X an Y coordinates in the point cloud are very good too. But in the Z are errors between 20 and about 50 cm at 11 different Controlpoints. The errors are not constant or homogeneous. Even the GCPs have this height-error (??).

UAV-complete-report.pdf (143.8 KB)

For my own control i have re-calculated one part of the project only with one group of photos (60m). Here are the X and Y very good too. Some Z-values are 1-cm- accurate, some have an offset of 10 cm at the controlpoints. In the point cloud i have seen, that below the surface is another “layer” of points in 2-3 m gap, but not everywhere (photo). What is this?

Can someone help with my issues? Have i done something wrong?

Überprüfung C 3 2 60m_report.pdf (1.5 MB)

[Blaz]

Hi @Saltcit, thank you for the detailed description and all the files!

I had a close look and could not see the checkpoints in the quality report for the whole project- did you perform the quality assessment in a third party software?

Looking at the absolute geolocation variance, the Sigma values are quite high. For sure higher that what we would expect for a RTK drone. Were there some issues with getting the GNSS fix during the flight?

As a side note, we developed a new desktop application Pix4Dmatic that is optimized for large scale projects. It is designed to handle and process thousands of images with ease while maintaining survey-grade accuracy.

Feel free to download it and try it out.

Best regards,

[Saltcity]

Hi @Blaz , thanks for your answer.

I have checked our Controlpoints within the Pointcloud and haven’t loaded them in the project.

We have controlpoints in the field with high-precise coordinates. So i just clicked them in the pointcloud and checked the z-values.

During the flight some pictures didn’t get fixed RTK, but this were only a few (<20).

I think the question is, were these points below the realpoint cloud in the picture above are coming from?

I think we don’t need pix4Dmatic so far, the control-project has only 853 photos an we just bought pix4Dmapper.

Do you have any other idea what can cause these errors or do you need some more data?

Thanks!

[Blaz]

The points that are below the dense point cloud could indicate noise or some calibration issues.

Could you click on some points in the point cloud and see how they reprojects back on images. Do the reproject back on the same location on images? If not, then this indicates calibration issues and you should place some MTPs in the project to assist the reconstruction.

Reprojection error

The area that you fly is challenging for reconstruction and this can also affect the accuracy. There are many similar visual features that are hard to be matched across different images. Could you try changing the following processing options and see if the results get better:

• Process with lower Keypoints Image Scale (for example, 1/2): This processing option can lead to a higher number of calibrated images than the default original keypoint image scale. For more information: Menu Process > Processing Options… > 1. Initial Processing > General.
• Enable Geometrically Verified Matching.

Let us know how it goes,