GCP shift after processing

Hi guys,

For this project, I have used 4 GCPs. I find that after processing, the GCP location (marked in black and white in the image below) has been shifted from the place it should be (the red cross location) by -0.9 in the X direction and by -0.4 in the Y direction… I don’t quite understand this… The red cross now has the coordinates of the manually entered black and white GCP coordinates, but in return, this black and white GCP is now shifted from its original location…

Why does this happen? What does it mean?

Does it mean my orthophoto and point cloud are inaccurate by -0.9 and -0.4 ?

Many thanks!

Hi Ahmad,

I am a bit confused by the fact that you have both a GCP photogrammetric target and a red cross. I strongly advise following our best practice instructions from this article:
Getting GCPs on the field or through other sources (optional but recommended)

I would also recommend the following:

  1. Please use more GCPs (5-8 for a start) which should be placed homogeneously in the area of interest. Imagine the area as a large table and the GCPs as the legs that will support it. If all the “legs” are placed at the same location of the “table”, then it will tilt. If the “legs” are homogeneously spread, then the “table” will be stable. Additionally, it is also recommended to place one GCP in the center of the area in order to further increase the quality of the reconstruction.

  2. Please note that the zoom level at which GCPs are marked has an impact on the GCP RMS error obtained in the Quality Report, so ensure to mark them at high zoom level. Usually, the higher the zoom level, the more precisely the GCP is marked. These GCPs will have more impact on the reconstructed model than GCPs marked in a lower zoom level and lower error values are expected for those GCPs. For example, when GCPs are marked without zooming into the images, the GCP error can be 10 times higher than when GCPs are marked by zooming into the images.

  3. Do not place the GCPs exactly at the edges of the area, as they will only be visible in a few images.

Cheers,
Teodora

1 Like

Hi Teodora,

Thank you for your reply. But it doesn’t really answer my specific question. It seems your answer is very general.

I used the (black and white) GCP as a system generated GCP once you capture these coordinates on-site and enter them manually to the system.

Then, once the drone flight is done, I tag the red gcps in the images to each corresponding (black and white) GCP location.

After processing, The coordinates of the center of the Red Cross are different than the coordinates of it when captured on-site by -0.9 in the X direction and by -0.4 in the Y direction… My question is, why did this shift happen?

Hi,

Let me emphasize the fact that 4 GCPs is not really enough to properly georeference a project. Secondly, I was not personally on the field to see how you measured these GCPs. Generally, a GCP photogrammetric target (figure below) should have about five to ten times the dimensions of the GSD. If the GCP is a natural feature (i.e. a characteristic point in the area that is not signed by a photogrammetric target), then the GCP can be even more difficult to identify and mark.

image

Please send me the quality report, so that I can have a better look.

Best,
Teodora

Please see attached.

Hi,

The RMS errors are quite small (~4 cm), which indicates quite a well-georeferenced 3D model.

image

If you really wish to improve the results, I recommend adding GCPs homogeneously across the dataset (please mark one also in the center). Because you added GCPs only at the extremities, some bias has been induced in the 3D model. The fact that your project deals with linear structures (i.e. streets) makes the georeferencing quality even worse.

Good luck!

Teodora

2 Likes

Thanks a lot Teodora :slight_smile:

1 Like