Phantom 4 rtk + d-RTK 2 base workflow. Z errors. Any suggestions please?

[ian_riverscience]

Hi Everyone
Sorry this is a long message. I've tried to give as much detail as possible to help with replies.
 
I’m stuck and any assistance would be much appreciated! I’m processing Phantom 4 rtk data collected whilst connected to a d-RTK 2 DJI base station in Metashape Professional 1.6.2 and getting large z errors in one of my data sets and can’t work out why or how / if it can be solved. Ultimately I am going to export the DSM into ArcGIS to calculate volume of soil loss in the gulleys that are present across the bare earth fields using an ArcGIS toolbox.

My normal workflow is to install a ground marker for the base station, survey its location with Trimble R10 with real-time corrections via VRS. The Trimble was set to survey in OSGB OSTN15. I look at coordinates of the point on the Trimble, after converting them to WGS84 lat long in decimal degrees (for the P4 rtk controller). Set up d-RTK 2 base station over the known point just surveyed. Put out markers and survey in their location with the Trimble R10 (still set as OSGB OSTN15). Start up P4 rtk and controller and base station, input the known coordinates into the controller, but first adding 1.802m to the z value to account for the base station DJI pole and save (using the settings button), fly with rtk on and connected to the base for corrections. Fly the pre-programmed flight path using 2D and/or 3D photogrammetry options.
Add photos into Metashape including XMP data, align photos, use gradual selection to remove some of the noise in the data.  Check box in camera calibration to enable rolling shutter compensation. Export coords of markers from Trimble in WGS84 lat long decimal degrees csv file. Import markers into Metashape, filter photos on each marker, one at a time and drag flag into centre to locate. Check reference settings coordinate system is WGS84 (EPSG::4326). Camera accuracy set at 0.03m. Marker accuracy set at 0.01m. All markers and cameras are left ticked for the following step. Optimise cameras using adaptive camera model fitting. This normally gives v good accuracy, e.g see the first attachment.

But with the problem data set, collected on the same day as the one above, over bare earth fields, I’m getting much larger errors in the z field. See the 2nd attachment – errors in the x and y for the gcps about 2cm, but 35cm in the z. I flew this at 2 altitudes, one using the 2D option (nadir only) at a height of ~30m, the other over a smaller subsection of the field with gullies using the 3D option (oblique) at a height of ~25m. I’ve tried with both sets separately and one combined file as well (which has 2175 images). I get these errors with all options.

If anyone has any suggestions on how I might be able to improve the z value, sources of such large error or whether this will be as good as it gets with this data set I would be very grateful. I’m just scratching my head why this data set is so poor with the z values when I can get good results at other sites with the same workflow. Many thanks in advance.

[ruadi]

Hello Ian_riverscience,

Yes I encountered similar problems using P4RTK with an NTRIP-Caster, but the RTK-datasource is not the problem. The main issue is the wrong f-value in the camera calibration. The auto-calibration is not capable to determine the +/- appriopriate value, all other values are meaningful (!).
Four our P4RTK, instead of 3666.666 (8.8mm/0.0024) we got 3678. For low flights (e.g. 40m) the error is not so severe, but flying at 120m amplifies the error.
How to get the correct f-value. Take an already flown campaign with 200 images and iterate the f-value (only this value!) by reprocessing, until you find our best f-value, actually to estimate  the true focal length for your device. A bit tedious, but it works. It is not clear for me how stable the estimated value is over longer periods (months), concerning temperature and mechanical movements.
Try it.

[ian_riverscience]

Thanks Ruadi. I appreciate the advice. I'll give that a try and see how I go on.

[Julien]

Hello

I agree with ruadi even if i dont do it this way. You need a reference mission with plenty of gcp. Do not import the calibration and you can let the rolling shutter compensation unchecked as you use the phantom 4 rtk (it is a mechanical shutter).
Perform the flight with the standart 3d option ie oblique pictures  and large margin auround the surveyed pitch.
First align with half of the gcp to verify the precision then realign with all the gcp, optimize.
Go in the adjusted calibration parameters to find your true parameters you can save to reuse them in your next projects.

[ian_riverscience]

Thanks Julien. I'll do that. 

[sfmdd]

You also can fly an oblique and a nadir mission on the same height (or the combined mission, multioriented) over the same area and process the datasets together. Metashape will estimate the correct focal length. At least this worked for me on many different spots without GCPs.

The estimated focal length for my P4RTK is around 3655-3656. Ive only flying missions with NTRIP over mobile network and no base station.

I got Z-Errors around +-2-5 cm depending on flight height to known points and airborne lidar datasets.

[Krishan]

Hello, we are also facing similar issue. How was the wind conditions when you got problem dataset?
Did it resolve after calibration?

[HansR]

Hi!
If you are using P4RTK the calibration from DJI is very good. BUT ONLY if you deactivate the DEWARP feature on the drone!!!!! (which you should if you do photogrammetry).
If so, the f,cx,cy,K1-K3,P1,P2 values are stored in *.jpg and automatically imported in Agisoft. You can set f and cx,cy to fix if you like.

Also, as mentioned earlier, oblique images assist a good bundle adjustment and camera calibration. Generally fly at 85° ant not 90° and you won't have troubles at all.

Hope this helps or at least give some ideas.
cheers

PS: The screenshot is an EXAMPLE of one drone. Please remind that all cameras are differnt. You can NOT adopt them to yours.