We have a project where 1) the circumstances make it prohibitively expensive to use GPS equipment to collect GCP coordinates, and 2) we have no need for georeferenced outputs.
We’re only going to be using the orthomosaic to draw linework, no elevations needed.
The plan is to collect the coordinates for the GCPs in an arbitrary coordinate system, but I see that the AutoGCP functionality does not support arbitrary systems.
I was thinking maybe I could download some georeferenced satellite imagery of our area of interest, and “transform” our arbitrary coordinates so that the GCP coordinates approximately line up with their real-world locations. My understanding is that the absolute accuracy of those “transformed” GCP coordinates would be terrible obviously (or unknown at best), but the relative accuracy should remain unchanged, correct?
The only thing I can think of is that the Quality Report might show some very high error values in the geolocation section, but I’m wondering if that actually matters when we aren’t concerned about georeferencing the outputs.
I’m planning on testing this idea using a georeferenced dataset that we already have good outputs for, and then comparing the results. I was going to take the GCP coordinates (collected with GPS) from this old job, and just shift/rotate the points slightly in CAD to simulate me “faking” projected coordinates, then export the “faked” coordinates and try to process the dataset again.
If anyone else has tried this before or actually knows enough to know that it won’t work, please let me know! Maybe I’m not seeing an obvious flaw in this plan lol. Thanks!
Hi @cmccurdy ,
Have you eventually converted you GCP’s coordinates and tested your idea? Do you want to share here some results?
What is the relative accuracy that you are looking for? Are your images geotagged?
Generally, one can expect an error of 1-3 times the Ground Sampling Distance (GSD) of the original images for the relative position of a point in a project that is correctly scaled and reconstructed. In case you have MTPs (GCPs), the relative accuracy can be increased.
Here you can read more about the expected accuracy of your project.
As you mentioned, the AutoGCPs functionality needs in input a coordinate system that cannot be arbitrary.
Hi Alice!
We aim for a relative accuracy of 0.10 ft. And yes, the images are geotagged (Phantom 4 Pro v2 non-RTK).
I tested my idea and I’m sad to say that it didn’t work. Soon after I received the email notifying me that the processing had begun, I received another email notification stating that “Detection of the GCPs failed” and that processing would proceed without GCPs.
However, even without the GCPs, the relative accuracy of the orthomosaic was indistinguishable from the control output (which had been previously processed with REAL GCP coordinates). The absolute accuracy was quite a bit off obviously, as it was georeferenced using only the GPS data in the image metadata. I think it was off by about 30-40 feet. I was able to correct the georeferencing using QGIS however, so that’s an easy fix.
I think if we go ahead with this future project, I will just process the dataset without GCP coordinates and georeference the orthomosaic after the fact in QGIS.
Thanks!
1 Like
Hello!
I understand. Then there might have been some issues with the coordinates (conversion) or the coordinate system.
As you say, if the relative accuracy you get is acceptable and you are not strictly interested in the absolute accuracy at this stage, processing without GCPs should also do the job for your case.
Thank you for sharing the information! I hope it will be helpful for other users passing from here too 