Support Website Contact Support Blog

Marks outlier detected for GCP tie points

As I am marking my GCP’s I will get several images marked to the GCP but then I will start getting the red indicator Outlier detected. How does this affect my GCP alignment and ultimately the Point Cloud and other deliverables. Can I leave them and the software will correct or do I need to remove those marks from the GCP images. I am using a Matrice 300 RTK with a P1 Camera flying Smart Oblique at 300 feet AGL at 14 MPH with a 70% frontal / 50% side overlap.

Hi @irondronecrew

Thanks for asking your question on the community. The answer to your question will depend on when you see the red indicators, but also on the quality of the calibration and the marks that were added. In general, the red indicators should not be ignored and should be removed by the steps described below.

Typical steps for including Ground Control Points (GCPs) in a project are:

  1. Import images
  2. Import GCP file & select the GCP coordinate system
  3. (optional) mark BEFORE the calibrate step
  4. Run the “calibrate” step
  5. (optional) mark AFTER the calibrate step
  6. (optional) If marked AFTER the calibrate step, run Reoptimize.

As you can see, the two main ways to include GCPs are:

  • Before the calibration (point 3 above)
  • After the calibration (point 5 and 6 above)

Now, in regards to red indicators in the tie points table:

BEFORE
If you mark the GCPs before calibration, the likelihood that there are red indicators is slightly higher, as the cameras positions & orientations were not “calibrated” yet, the alignment between them might not be great yet. This means, that the software is likely to interpret a mark as an outlier.

Two things to do in this case:

  1. Check your marks and make sure they are at the right spot on the images. If you are sure the marks are at the right spot in the images, then it is likely the outlier marks will go away once the calibration is done, as the calibration will align all images correctly and will then not see a “misalignment” between the marks.

  2. Run the calibration, if the outliers are still there, either the calibration did not work well, or there is still a mark that is not at the right spot in the images. In that case, I would check the position and reprojection errors in the tie points table for the different GCPs, there you should be able to spot which have an unusually high error. It’s easy to order them by decreasing size with the little arrows in each column. An outlier would typically be translated into a high reprojection error, e.g.

    in that case, it is recommended to check your marks again, remove or improve the ones that are not accurate enough. This should improve the reprojection errors that is shown, if that’s the case, you can Reoptimize by selecting the “Calibrate” step and the “reoptimize cameras” toggle within that.

AFTER

If you mark the GCPs after calibration and the calibration seems OK, then it is likely you made a marking mistake. In that case, same tips as in point 2. above.

VERIFYING MARKS - TIPS

A few tips to make it easier to identify outliers in your project:

  • CTRL+F or “…” > “Maximize image viewer”, get your images fullscreen and adjust the number of images you see at once. This helps to recognize patterns in marks that are not correct. Example:

  • Hit “space” when hovering an image gets you a fullscreen view of an image. You can zoom in/out and go to the next image with the arrow keys. This can be accessed from the normal image viewer or from the maximized image viewer e.g.

*Sort images by distance, reprojection error or alphabetical order. Sorting by reprojection error typically helps to identify outliers, in the example above it sets the outlier directly at the top. This can be accessed in the standard or maximized image viewers, e.g.

  • The Tie Points table is a good place to look for signs too, I’m listing it here, but an example was provided higher up in the reply.

  • Ensure that all marks are exactly at the same place, a tie point is an input from the user to the software. The software trusts that input, so you need to make sure it is accurate. To ensure you mark at the same spot, I would always recommend to measure GCPs on features that have a high contrast and that can be easily spotted in images. When marking, ensure that you zoom in on the spot to ensure an accurate mark.

Those are a few tips that should help fix such issues. Note that we are considering adding pink circles directly in the images that are considered as outliers to facilitate this type of work in the future. The answer ended up being longer than I thought, but I hope it helps!

1 Like

Great info!!