Accuracy of DSM with nadir images

PIX4Dmapper PIX4Dmapper Questions/Troubleshooting
1

I had a problem with my last survey (about 300 nadir images, GSD 2.66cm, flight height 60m and 8GCP). 

From quality report:

Mean [m] -0.033241 0.044439 0.049926
Sigma [m] 0.095614 0.124330 0.116952
RMS Error [m] 0.101227 0.132033 0.127163

However the output has some issues.

Mosaic result is very good, I don’t notice remarkable differences between mosaic and official maps but I compared quotes from DTM/DSM with official marks and our survey done with GPS/Total station and the difference is generally about 50cm.

Below the list of GCPs: 

GCP Name Accuracy XY/Z [m] Error X [m] Error Y [m] Error Z [m] Projection Error [pixel] Verified/Marked
GCP1 (3D) 0.005/ 0.007 -0.266 0.354 0.354 1.127 15 / 15
GCP2 (3D) 0.006/ 0.009 0.004 0.029 0.004 0.976 11 / 11
GCP3 (3D) 0.011/ 0.016 -0.010 0.073 0.025 1.599 11 / 12
GCP4 (3D) 0.010/ 0.015 -0.072 0.036 0.033 1.018 8 / 8
GCP5 (3D) 0.010/ 0.017 -0.027 -0.007 -0.003 1.297 6 / 6
GCP6 (3D) 0.010/ 0.018 0.022 -0.039 0.033 0.543 8 / 8
GCP7 (3D) 0.009/ 0.016 0.014 -0.026 -0.028 0.993 13 / 13
GCP8 (3D) 0.015/ 0.028 0.069 -0.066 -0.018 0.865 10 / 10

GCP1 also looks strange because it’s the most visible and clear and his place is close to the border and not in the center 

It’s first that I have a gap so important. Is there any reason and I can fix this issue?

Thanks.

 

2

Could you provide the entire quality report of this project? For example with a Google Drive or Dropbox link. This could help to solve the issue. 

The first I noticed was that the accuracy that is defined for the GCPs is very low e.g. 5mm/7mm for GCP1. The default value of the GCPs’ accuracy is 2cm. As the average GSD is about 2.66cm, it will not be possible to mark them more accurately than that. This is why I would suggest to set the accuracy to the default value of 2cm. This should reduce the constraints on the project by the GCPs, which should result in a better reconstruction. Process > Reoptimize should be sufficient once the accuracy is edited. Here is how you can edit it: https://support.pix4d.com/hc/en-us/articles/202557919#label14

I suggest to make a duplicate of your project with Project > Save Project As…, so that you can safely try these suggestions. Let us know if that could improve the project. Feel free to post the quality report of the original project and of the one with the suggestions applied. 

3

Hi Pierangelo,

thanks for the support. I changed the parameters and redo step 2 and 3 after reoptimize. Now it looks better but far for having a good result.

Please find below the quality report:

https://www.dropbox.com/s/v4rphrcalc3me49/1289_001_Tognano_20170913_report.pdf?dl=0

Other jobs were good so It should be an error somewhere during the survey or the process 

I also send a small check of some points: some are cadastral point and some are GPS points taken by us. I compared my result with swissalti and I always get a bigger difference. 

https://www.dropbox.com/s/yh4yvu3ukimxtlz/drone%20tognano.pdf?dl=0

 

 

4

Hi Massimiliano, 

Thanks for attaching the quality report. As for the additional document you sent, I have some difficulties to understand what each measure corresponds to in the screenshots. Could you explain what layer corresponds to swisstopo, to your reference measurement and to the Digital Terrain Model (DTM) from Pix4D? Please also mention the units for the differences you have noticed. 

Below is the table of GCPs in the quality report with the ones that have a large error in yellow: 

There are three GCPs with a larger error, I cannot see where these are placed in the project area, but my guess would be that these are the ones on the border of the project: 

The reason is that there is a lower overlap, which could compromise the accuracy of the reconstruction in these areas: 

Can you tell us where the GCPs with the errors are placed? 

Here is an article with the recommended GCP distribution in a project: https://support.pix4d.com/hc/en-us/articles/202557489 

Note that for creating DTMs there is a new point cloud classification feature in version 4.0 that improves the creation of the DTM, you might want to try that.