Author Topic: Discrepancy between GNSS-measured GCPs and Drohne DJI Mavic E3 RTK (Read 13512 times)

Photo · « **on:** October 23, 2023, 10:47:02 AM »

Hello everyone,

i recently conducted a survey using the DJI Mavic Enterprise 3 RTK and subsequently measured several Ground Control Points (GCPs) using a GNSS antenna. Both devices, the drone and the GNSS antenna, have an accuracy of less than 2 cm, which I have already verified.

However, after importing the data into Metashape, I notice a discrepancy of about 40 cm between the Drohne and the GNSS-measured points. All data was captured in WGS 84, and no transformations were made.

Here are some points I've already checked:

The accuracy of the GNSS antenna and the drone.
The RTK connection during the flight was stable.

I'm uncertain about the source of the issue and would greatly appreciate any guidance or suggestions. Has anyone experienced a similar problem or have ideas about the potential cause?

Thank you in advance!

Best regards,
Photo

SFL_Beda · « **Reply #1 on:** October 23, 2023, 11:13:23 AM »

What was the RTK reference? Was it the same for both devices and did you do it at the same time?

How did you verify the 2cm accuracy? (In my experiments the Mavic3 Enterprise was more like 10cm accurate)

If you say accuracy do you mean relative or absolute accuracy?

Photo · « **Reply #2 on:** October 23, 2023, 11:51:54 AM »

Hello SFL-Beda,

Thank you for your response. If I understand your questions correctly, the drone has an RTK module and everything is measured in WGS84, and the GNSS antenna does the same. Both of these devices use the same service (referring here to the correction data).

To answer your second question, according to the manufacturer's specifications for the drone, the accuracy is 1.0cm horizontal and 1.5cm vertical. The accuracy of the GNSS antenna varies between 1 cm and 3 cm.

I can see the accuracies of my images in the EXIF data, and almost all of them were around 2 cm, and the measured GCPs were also at 2 cm.

When I calculate an alignment in Metashape and then import my points, the points are about 40 cm off from the marker.

To answer your last question, I am referring to absolute accuracy.

SFL_Beda · « **Reply #3 on:** October 23, 2023, 12:18:01 PM »

So if I read it correctly you didn't actually "verify" the accuracy of anything, you just trust the values given by the manufacturer.

Those are usually best case scenarios and the values in the exif data are often a bit overoptimistic.
(also for the drone the spec is "Horizontal: 1 cm + 1 ppm; Vertical: 1.5 cm + 1 ppm", if you use a network RTK correction and are over 10km away from the nearest station, the ppm error will actually become dominant).

The same is true for the GPS Rover and that one is very sensible to user error (you need to keep the pole exactly straight, it's really easy to add 1-2cm of error by not leveling the pole correctly).

If you get a offset of 40cm though that would seem a bit too much, especially if it is consistent over multiple ground control points. Can you check what direction the error occurs in? If it is in z-direction, my guess would be that you got the pole offset of the GPS rover wrong.
I assume that you use the same height reference as well?
The drone uses above ellipsoid normally, if the rover is set to above geoid, that could also lead to a static height offset.

Photo · « **Reply #4 on:** October 23, 2023, 12:46:59 PM »

Yes, the values provided by the manufacturers, as you described, are best-case scenarios. If I had a deviation of 4 or even up to 6 cm, I could live with it and it would be acceptable. But 40 cm is way too much.

Yes, I'm using the same height reference in both devices.

I've read about this, and someone once mentioned that one always has to correct the coordinates of the pictures by simply subtracting the distance to the nearest point. I read about this here, here's the link:
https://www.agisoft.com/forum/index.php?topic=13451.0
And this makes little sense to me. Why would I have a drone that has a centimeter-level accuracy, and then everything suddenly falls in the decimeter range?

The error direction is in the position, while the height, I wouldn't say it's not 100% accurate, but it's very minimal, and that's okay. Maybe it's off by 3 cm in height, but not more. The main issue is with the X and Y positions, in both directions.

SFL_Beda · « **Reply #5 on:** October 23, 2023, 01:09:11 PM »

They are talking about a setup with a mobile RTK base with unknown absolute location.
If you used the same network RTK for corrections this does not affect you at all.

From what you described, the measurements should be indeed correct but as they don't align, something must have gone wrong somewhere.

Photo · « **Reply #6 on:** October 23, 2023, 01:15:31 PM »

When I fly, I can see that my drone successfully connects with RTK and I can view its accuracy, which is always within the centimeter range. The same is true for the GNSS antenna. I currently don't understand the issue, as it seems many others have the same problem. Some people say they primarily use GCPs and therefore reference their model based on their GCPs. Sure, that's a valid approach and I fully support it. But then, why have an RTK drone that supposedly achieves measurements in the centimeter range?

Could something have been lost during the alignment in Metashape? I've checked everything possible, and camera calibration isn't the problem. Metashape handles that very well.

Alexey Pasumansky · « **Reply #7 on:** October 25, 2023, 05:58:33 PM »

Hello Photo,

Maybe you can share the project with alignment results (tie points, aligned cameras, markers) to support@agisoft.com, so that we could take a look at it?

Photo · « **Reply #8 on:** October 25, 2023, 06:53:40 PM »

Hello Alexey,

I just did this a couple of seconds ago. I hope to hear from you guys soon.

Best Regards,
Photo

ckurtze · « **Reply #9 on:** January 16, 2024, 04:46:42 PM »

Hi all,

I processed a flight mission made with an DJI Mavic 3 Enterprise with RTK module including some GCP as proof for accuracy. I found out, that the horizontal accuracy is quiet well (about 2 - 3 cm per picturecoordinate with reflects the accuracy of the RTK measurements) but the vertical accuracy is about 9 cm! Could it be, that Metashape doesn't use the offset betwenn GNSS antenna phase centre and camera with is written by the copter to the MRK file?

Best regards,
Christian

4DS Niklas · « **Reply #10 on:** January 17, 2024, 06:37:35 PM »

Hey everyone,

i have had similar experiences with RTK drones. Especially the error in elevation between the tie points and GCPs pops up every now and again.

I am not a technical expert, but we assumed it has to do with the camera calibration parameters, especially f, cannot be estimated with 100% certainty with just RTK positions without the additional info GCPs provide to the bundle adjustment. I would be very happy to hear about this from someone with more technical knowledge.

I also find that the RTK Positions from the Mavic 3E are not of the same quality as the P4 RTK. When using the P4, the software we used stopped the drone for every image taken. The Mavic 3E with the preinstalled DJI Pilot app does not allow this. This makes the flight faster, but i feel the positioning accuracy suffers because of this. The Drone itsfelf always estimates its accuracy to be way higher than the survey results with GCPs show in the end. Did anyone else have this experience?

Dieter · « **Reply #11 on:** January 17, 2024, 07:08:33 PM »

There are always problems with flat surfaces and a fixed flight altitude.
The camera calibration can't compensate for this accurately enough. This can be remedied by another overflight at different altitudes.

For example, the normal overflight altitude is 50 meters, then I will definitely fly again at a height of 80 meters. Then the evaluation will also work correctly.

There is enough research on the subject in i-net.

In addition, I can only recommend making a new camera calibration for each battery especially when there are temperature differences between flights.
And always "apply the additional corrections" during overflights when optimizing.

Dieter

Dieter · « **Reply #12 on:** January 17, 2024, 07:14:37 PM »

Take a look:

https://www.researchgate.net/publication/350556530_Photogrammetry_using_UAV-mounted_GNSS_RTK_Georeferencing_strategies_without_GCPs

Dieter

ckurtze · « **Reply #13 on:** January 18, 2024, 05:03:30 PM »

Thank you Dieter for your informations (can I contact you via phone?). The publications seems to be very interesting even though I am new in RTK flights. I usually used GCPs for georeferencing. But again my question more specific: Does the Mavic 3E apply the corrections written in the MRK-file already to the coordinates in the image file metadata? If not - do Metashape consider this correction data from the MRK-file?

Cheers,
Christian

Dieter · « **Reply #14 on:** January 18, 2024, 05:36:23 PM »

I had the Mavic 2 with the Topodrone RTK extension for 3 years and since the beginning of the year the Mavic 3E with RTK.
According to my experience and measurements, the RTK coordinate, which is stored in the image, refers to the center of the camera, so everything is correct.

I compared the height of the camera center point to a leveled point and the result fit to 1 cm.

Anything else wouldn't make sense

Dieter

Forum