Interpretation of GCPs and CPs errors to assess the quality of a project

Hi,

I have developed several projects with Pix4Dmapper for my degree thesis. One of the main issues is evaluating the quality of the results and comparing them with each other. I would like to be able to correctly interpret the tables with the errors of the GCPs and Check Points present in the quality report in order to do so.

Reading the various articles on the support site and several scientific publications and books I came to the conclusions written below, but I’m not sure they are correct. Can anyone tell me if it is all correct and if there are other ways of interpreting these values? Within what threshold are the errors acceptable and how, by looking at the error values, can I tell if one project is more accurate than another?

 

In general:

  • the lower RMS Error values are and the better;
  • if the mean errors are small (less than 25% of the RMSE values), it can be assumed that there is no systematic error in georeferencing;
  • the projection error should be lower than or equal to 1 pixel, the lower the better.

Regarding the GCPs errors in particular:

  • the errors, in absolute value, of each point must be equal to or less than the accuracy set for that direction.

Regarding the CPs errors:

  • the errors, in absolute value, of each point in horizontal direction (X or Y) should be less than or equal to 2*GSD;
  • the errors, in absolute value, of each point in vertical direction (Z) should be less than or equal to the 3*GSD.

 

Thank you all,

Chiara

 

This is the table of one of the projects realized, the GSD in this case is 4.20 cm.

Hey Chiara,

The lower the RMS error value the better it is, I am not sure about the less than 25% mean errors.Your GSD is 4.20 cm so projection error less than or equal to one pixel seem pretty accurate.For each point in horizontal direction and in vertical direction  I am not too sure what error should be. It really depends on what accuracy you are trying achieve, if you are trying for sub-centimeter accuracy then you need them to be more accurate. But if you are doing agriculture projects for example sub 10 centimeters would be very accurate. I think your GCP numbers look really good.