GCP Issues, output is tiny-scaled with GCP

I have two sets of images (~60) of the same terrain taken a month apart, and both process perfectly without GCP. They are within about 1-2m alignment of one another with GCP.

I’m adding GCPs in order to better align the images. Note this is using ODM directly and not WebODM, etc. In my first attempt, per the ODM documentation, I specified NaN for the elevation (geo_z). This results in an exception being thrown during the build process.

So now I’m specifying the elevations of the GCPs in meters above sea level. This is resulting in my output being very, very small. Like 1/10th scale of what it should be. The following excerpt from the processing summary illustrates this.

This is in comparison to processing without GCP which works fine:
(as a new user I can only embed one image, maybe I can post it as a follow up reply).

I have tried creating a gcp file in two different coordinate systems (“+proj=utm +zone=17S +ellps=WGS84 +datum=WGS84 +units=m +no_defs” and EPSG:4326) with the appropriate converted coordinates, and both result in the exact same output. I have triple-checked all GCP coordinates in Google Earth / Google Maps and they all appear to be correct. All GCPs are on the ground, and they are within 3 meters elevation of one another. GCP follows:

+proj=utm +zone=17 +ellps=WGS84 +datum=WGS84 +units=m +no_defs
490814.48 4088278.08 715 3191 1208 DJI_0872.JPG GW_SW
490877.93 4088399.73 716 1850 1080 DJI_0832.JPG GW_NW
490913.54 4088395.68 716 2030 1146 DJI_0830.JPG GW_NE
490954.28 4088324.46 717 1685 1117 DJI_0845.JPG GW_SE
490976.24 4088274.21 714 2495 1972 DJI_0879.JPG GW_TB

What is causing this strange scaling?

Good processing without GCP:


You need to tag each GCP in at least 3 images.

Thank you - I wish the documentation stated that. It says “It’s important that you find high-contrast objects that are found in at least 3 photos, and that you find a minimum of 5 objects.”

It doesn’t say they need to be tagged in 3 photos, but that they are visible in 3 photos. I assumed that a GCP tagged in one image would then be referenced visually in whatever images also contain that reconstructed point / feature, as visually identified in the processing.

Ah yes, the wording there could be improved a bit (help us improve the docs?) GitHub - OpenDroneMap/docs: 🎉 Contribute to OpenDroneMap's documentation! Read how below! 🎉

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.