Absolute accuracy with GCPs

Hello,

I’m using an Emlid Reach RS+ base and rover set to NAD83 2011 per my state’s NTRIP caster to shoot GCPs. I’ve done all the trouble shooting in the field that I can think to do and I’m still getting absolute accuracy of .3 m or so horizontally and .01 vertically when I process. My most recent effort was shot in the Ohio South coordinate system. I was able to input the proj4 code and images processed without error. I’m getting a consistent distortion pattern on my five points. If the base is located somewhere near the middle of the field, the four corner points are pulled toward the base, vertically, and pushed outward from the base horizontally. My center point, which is located closest to the base, is about dead on each time. I’ve read through all of the documentation pages and I can’t find a solution to this problem.

1 Like

Welcome!

Sorry you’ve been having trouble, but I thank you for taking the time to experiment and document what you’re seeing! Much appreciated!

Can you share the data and your processing parameters?

Gladly!
Proj.4 string is taken from EPSG.io:
+proj=lcc +lat_1=40.03333333333333 +lat_2=38.73333333333333 +lat_0=38 +lon_0=-82.5 +x_0=600000 +y_0=0 +ellps=GRS80 +units=m +no_defs

I’m using the webODM Lightning dashboard app. I create a new project, select the images and the GCP file. The GCP file is created with your GCP Editor Pro. Let me show you the pictures of the points from the GIS software. My base station is located closest to point 3. As you can see, point 3 is within the acceptable margin of error. Points 1 and 2 are to the north of the base station and positioned to the west and the east respectively. Points 4 and 5 are to the south of the base station and to the east and west respectively. I also get this kind of distortion no matter what my GNSS settings on the units. Do you know what’s causing this?

Point 1

1 Like

I have noticed the same thing. Playing with camera settings I’ve concluded that camera calibration can be a factor.

This page can give some answers:

1 Like

Is there someone who is getting good results able to put a workflow together? I’m happy to put my entire equipment list and workflow up if it would help resolve this issue.

I’ll also dig into camera calibration.

2 Likes

I do not know, but I may have some ideas…

Are you able to share the input images, GCPs, etc with us/me? If so, please post a link here or PM me.

I want to try some stuff.

1 Like

Something to play with: Uncertainty Priors for GCPs and Shot Position in Bundle Adjustment · Issue #590 · mapillary/OpenSfM · GitHub

1 Like

There’s the link to the images, the GCP file, and the survey file from my Emlid unit.

I’m running a Mavic 2 Pro on Drone Deploy’s capture app. Ohio’s NTRIP service is based on NAD83 (2011) and NAVD88. I have those selected in the Reachview app for the survey. Please let me know if you need any additional information and thank you very much for your help!

2 Likes

I would start by reducing the number of images where each GCP is tagged; 16+ images per GCP might be overkill. Usually 3-5 are sufficient.

3 Likes

Are you able to share your Report.PDF that you got from your processing?

Hey, sorry for the delay on this. The processing report has been added to the onedrive folder.

1 Like

No problem!

I just like knowing what I’m comparing my output to, and it helps me test a theory about constraining the --gps-accuracy flag with the results of a prior processing run.

Thank you for taking the time to add more data/info for us!

1 Like

My pleasure. I appreciate your help and I’m real interested to see what you come up with. Let me know if you need anything else.

2 Likes

Mm, don’t think I came up with much, unfortunately.

However, it looks like the ORB feature matching I implemented as a test might (significantly) reduce the error of the reconstruction.

That being said, ORB is really touchy right now and can fail to reconstruct many datasets that SIFT/HAHOG have no issues with.

skyview_8782_original_03m_1.pdf (3.3 MB)
image
image

skyview_8782_preprocessed_ORB_1.pdf (3.6 MB)
image
image

So is it something I’m doing wrong in the field?

I’m not certain, unfortunately. Seems likely not.

What angle are you capturing at? Nadir? Try a few degrees off.

I can’t seem to be able to download the files (asks for a login)?

1 Like

It is nadir capture.

1 Like

Give that a try.

Worked, thanks!

1 Like