Forum

Author Topic: Geometry Accuracy Calculations/Validation  (Read 9619 times)

tdepke

  • Newbie
  • *
  • Posts: 21
    • View Profile
    • Graduate Student Profile
Geometry Accuracy Calculations/Validation
« on: November 19, 2011, 05:52:07 AM »
Howdy all,

Does anyone know if there is a "best" method for calculating or estimating the overall accuracy of a model that is generated? From all that I have read the only way to do that is with comparing actual elevation values that are surveyed independently or determined from a different method. Lets say I only have 15 GCPs that each have their own RSME value, is there anyway to estimate the accuracy across the model into a single value from the 15 RMSE values that I have? I haven't found anything in the literature and I just wanted to see if anyone knew.

Thanks,

Tyler

tdepke

  • Newbie
  • *
  • Posts: 21
    • View Profile
    • Graduate Student Profile
Re: Geometry Accuracy Calculations/Validation
« Reply #1 on: November 19, 2011, 09:33:19 AM »
I just saw that in the GCP dialogue there is a TOTAL at the bottom that includes a Total RMSE. How is this value calculated? I exported all my RMSE values to Excel and did an average of RMSE values and although it was close, it was not the same as the TOTAL value.

Thanks,

Tyler

tdepke

  • Newbie
  • *
  • Posts: 21
    • View Profile
    • Graduate Student Profile
Re: Geometry Accuracy Calculations/Validation
« Reply #2 on: November 21, 2011, 02:03:05 PM »
Howdy again,

I am still looking for answers on these previous two posts...anyone know anything regarding these questions?

Also, regarding checking and unchecking GCPs in the GCP dialogue. What does unchecking a previously checked GCP actually do to an already created DTM? Usually I've been unchecking the GCP's that have really high RMSE values, which lowers the other RMSE values, but I would like a clear explanation as to what this actually does.

Thanks,

Tyler

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14816
    • View Profile
Re: Geometry Accuracy Calculations/Validation
« Reply #3 on: November 21, 2011, 05:50:29 PM »
Hello Tyler,

Total error in Ground Control dialog box is calculated as square root from the sum of squares and that's all is divided by the number of GCPs. Maybe the difference you've got could be caused by division by (n-1).

As for the unchecked GCPs - their coordinates are not taken into account when coordinates of the model are assigned. You can see the change of the errors clicking Update button after checking/unchecking GCP.
But if you are sure in coordinates and accuracy of the GCPs but the error is still high that usually mean that the model has deformations which could be compensated by using Optimize feature (in 0.8.4).
Best regards,
Alexey Pasumansky,
Agisoft LLC

tdepke

  • Newbie
  • *
  • Posts: 21
    • View Profile
    • Graduate Student Profile
Re: Geometry Accuracy Calculations/Validation
« Reply #4 on: November 23, 2011, 10:44:55 AM »
Thanks for the clarification Alexey. I really appreciate it.

tdepke

  • Newbie
  • *
  • Posts: 21
    • View Profile
    • Graduate Student Profile
Re: Geometry Accuracy Calculations/Validation
« Reply #5 on: November 23, 2011, 01:34:11 PM »
I don't know if other people have requested this for future versions, but I think that including the RMSE values in both pixel and distance (m) would be beneficial for users. Maybe some kind of toggle option or setting where a user could change units px, m, ft, etc.

gEEvEE

  • Jr. Member
  • **
  • Posts: 66
    • View Profile
Re: Geometry Accuracy Calculations/Validation
« Reply #6 on: November 29, 2011, 04:59:51 PM »
Hi,

I did quite some validations of the PS output. Many things are still under review: one article is already published here: http://cipa.icomos.org/text%20files/PRAGA/Doneus.pdf
Bear in mind that is the conference proceeding. There will be a slightly improved version coming out later (we needed to corrects some issues with filtering the laser data, but the conclusions are the same).

Another article will be published in Archaeometry (it is accepted at this very moment). Anyhow, I will post a note when it is available. Most of my papers can be found here: http://lbi.academia.edu/GeertVerhoeven. It will upload the Archaeometry text once it is published and the newer version of the first paper I mentioned (also when it is published).

Cheers,

Geert