Aligning photos with Ground Control Points

[Krister]

Hi!
I am a surveyor that have used Photoscan for a couple of years. I just recently updated Photoscan to Metashape and am wondering about the Align photos section.

In Photoscan I unchecked all Cameras and turned off Generic preselection since the drones GPS only have about 10m in accuracy and volume measurement always gets wrong.
Instead I used N-RTK measured GCPs (ground control points) That makes the accuracy to a couple of centimeters and enabled reference preselection in the Align photos section. And that always worked well.

My question is what is the best way of working order now in Metashape? What preselections works best with GCPs and why?

I would be greatful for some advice!

[Ketzerisch]

Hey Krister,

it sounds a bit like you are mixing some things up. First of all your GCPs have nothing to do with the alignment process. During the alignment process Metashape matches the pictures based on equal points in mutlple pictures. In the generic preselection Metashape downscales your pictures and searches for pictures which were taken from neighboured positions. When you have approximiations for position and angle of camera (like you have in the EXIF of your drone photos) you can select preference preselection so Metashape already knows which picture is next to which an do the aligning process faster. So you always want to chose reference preselection with your drone data.

The GCPs come only into play after the aligning process: You mark them in the aligned photos to convert in the desired coordinate systeme, optimize the camera locations and estimate your camera calibration.

[Krister]

Thanks for your reply!
I wonder if we talked around each other, but to be sure I will precise how I worked in photoscan.

-After the flight I created a new project in Photoscan.
-Imported the photos into a new chunk.
-Convert photo coordinates from WGS84 to wished coordinate system.
-Align photos with generic preselection on lowest accuracy and 80 000 key point limit and 8 000 tie point limit.
-Optimize Alignment.
-Import GCP coordinates.
-Place the markers on the photos with the corresponding GCPs.
-Uncheck Photos in the reference tab.
-Align photos again, but now on the highest accuracy, turn of generic reference preselection, turn on reference preselection, 80 000 key point limit and 8 000 tie point limit.
-Optimize Alignment.
-Build dense cloud
-Build Mesh
-Build DEM
-Build orthomosaic etc.

Here is the generated report from the work processed in the old Photoscan:









Here is the difference between N-RTK measured Control Points (made in the same way as the GCPs) and Interpreted center of the Control Points in Photoscan Orthomosaic.


I used this method because volumes always went wrong when mixing generic reference preselction with reference preselection (compared with traditional GPS measuring)
With the method I've been learned in Photoscan the volumes always was within about 3% compared to traditional GPS measuring).

I see that people are working in different ways and I would gladly know the difference in speed and accuracy. And please tell me if I have misunderstood anything and why.