Agisoft Metashape

Agisoft Metashape => General => Topic started by: sdesimone on November 02, 2020, 02:54:12 PM

Title: Discrete measurements of checkpoints in Metashape
Post by: sdesimone on November 02, 2020, 02:54:12 PM
Good morning,

I am writing regarding checkpoints in Metashape. I wish to take discrete measurements of checkpoints in Metashape and I have encountered some issues during the placing of one point.

I have taken three different set of photographs of the same subject using:
- A mixed geometry photogrammetry (in which I have convergent photogrammetry, near-parallel photogrammetry and close-ups of the subject);
- A convergent photogrammetry;
- A near parallel photogrammetry.

I have the same set of 14 Ground Control Points (coordinates taken with a robotic total station in British National Grids coordinates) and 27 checkpoints (coordinates taken with a robotic total station in British National Grids coordinates).
Therefore, I have created three different projects in Metashape for the Mixed, Convergent and Near-Parallel photogrammetry. I did two approaches to take discrete measurements of the checkpoints in all three photogrammetric approaches.

1) Importing the photographs, aligning and importing in the "Marker List" the coordinates for both the Ground control points (GCPs) and checkpoints. I unchecked the checkpoints and placed all the GCPs in my photos. I did the first bundle-adjustment and checked the coordinates for the checkpoint in the marker list. At this point I can see the position of my checkpoints in the set of photos. I checked their position and moved "the marker" when necessary (the flag then turned from blue to green in the photos). Once done I have re unchecked the coordinates in the "Marker list" and did a second bundle adjustment.
In this way I have an error (m) for both each individual checkpoint and an overall error (m). I have then filtered the point cloud, run a third bundle adjustment and finally set the dense point cloud to run.
This approach has been fine with all three different photogrammetric strategies. My issue arise with the second approach.

I wanted to compare the coordinates of the checkpoints taken with the total station (the existent set of coordinates that I have) with coordinates given by placing brand-new-markers by me in Agisoft.
For this approach the steps have been the following:
1) importing the photographs in metashape and aligning them. I have then imported the GCPs and placed them all around my sent of photos. Once done I have done a bundle-adjustment I placed the checkpoints as "new marker" in the set of photos, adjusting their position in all the photographs. When I needed to do this procedure with the Convergent photogrammetry one point was visibly shifted from the first picture to the second one and so on. Therefore, it gave me a much larger error (14 cm compared to few mm I had in all the others). Even if I adjust its position in all the photos the error remain large.
I have not encountered this issue with the Near Parallel photogrammetry strategy, in which all the points could be placed with no issues.

I would like to then ask a couple of questions:
1) If you have any suggestion why this error could be happening? especially why only that specific point in that photogrammetric approach.

2) I would also like to ask if the overall error (m) for the checkpoints represent a a full-3D circular error estimate? How's that value calculated?

Thank you very much in advance for any help in the matter.
Best wishes,
Samantha De Simone
Title: Re: Discrete measurements of checkpoints in Metashape
Post by: JMR on November 08, 2020, 09:03:18 PM
Hello, Samantha:
I cant fully explain your error without further analysing your dataset and fully understanding your steps. when it comes tho the las question, the 3d error of a check point corresponds to the differences between input and estimated coordinates and to give this error in a single number they are just summarized as a quadratic mean

Title: Re: Discrete measurements of checkpoints in Metashape
Post by: sdesimone on November 09, 2020, 12:22:06 PM
Dear JMR,

Thank you very much for the reply. I understand without the dataset to look at the first question is tricky, but thank you for your reply to my last one, much appreciated.