Interpreting RMS error with GCP's

Hello everyone 

      I’m having an issue understanding the RMS error given in the Quality Report. My project has a GSD of ~5cm/pixels, 6 3D GCP’s marked on the RayClould and after processing it’s showing a mean RMS error of 0,003m.

     From my understanding the minimum error should be half of the planimetric resolution, in this case is 5cm, so it should be 2.5cm, and I should expect and error of 1-3 times the GSD. But the quality report shows 3mm.  

      I know the project is really accurate and the software does an amazing job analyzing all data and I’m just trying to get more information on the values of the errors and what my real error is.

Thank you

Hello Bruno,

The RMS error of the GCPs corresponds to the difference between the initial and computed positions of the GCPs. It is a local indicator of how well Pix4Dmapper fitted the model to the GCPs.

These geolocation errors do not assess the accuracy of the computed 3D points, i.e. the automatic tie points that make the model. They assess the consistency between the geolocation measured on the field and the geolocation as estimated by Pix4Dmapper when processing the model.

On one hand, our software tries to adjust the reconstruction using the GCPs meaning that the goal is to minimize the RMS error. This is why you ended up with 3 mm. On the other hand, the global absolute accuracy of the outputs should be closed to 1-2 pixels (GSD) horizontally (XY coordinates) and 2-3 pixels (GSD) vertically (Z coordinate) using GCPs. In your case the error wouldn’t exceed 10-15 cm.

More information on GCPs RMS error: https://support.pix4d.com/hc/en-us/articles/203604125

About accuracy of the outputs (absolute vs. relative): https://support.pix4d.com/hc/en-us/articles/202558889

Best regards,

 where can i find the vertical accuracy between the GCP’s that i manually entered? How do i know if my data is within a tenth vertically?

Hi Scott,

You can check the quality report and the RMS error as mentioned in the comment above to assess the GCP adjustment.
More generally, you can refer to this article to analyze your results: https://support.pix4d.com/hc/en-us/articles/202558689

Regards,

Hi Julie,

I am having some questions about the accuracy of my results and the meaning of the rms results. I am working on a topo project and centimetric accuracy is required.

On my fisrt area I have a 1.9 cm gsm 9 hectare of surface and three gcp’s. Result rms is 0.015, which correspond to 1.5 cm, does that mean my ùosaic will be 1.5 cm of accuracy ?

Geolocation Bias :
x : 0.319

y : 0.651

z : -1.95

in advance thanks,

John

Hi John,

The Ground Sampling Distance (GSD) is one of the parameters that can be used when evaluating the final accuracy of the project. In general, we say that the final results have an error of:

  • (Relative accuracy) 1-3 GSD.
  • (Absolute accuracy) 1-2 GSD horizontally (X,Y coordinates), 1-3 GSD vertically (Z coordinate).

In your case (1.9 cm GSD), we would say that the expected accuracy of the final results is between 1.9cm and 3.8cm. Keep in mind that the accuracy can vary depending on the area in the project, for example, areas, where there is a lot of dense vegetation, might have lower accuracy in the end. 

I would recommend measuring more point in the area in order to get a better overview of the situation. Some points can be used as checkpoints in order to get an independent estimation of accuracy. 

More about the error values in the quality report can be found in How are the GCP Errors defined in the Quality Report? - RMSarticle.

The Geolocation Bias is displayed only if GCPs are used in the project. It defines the bias between image initial and computed geolocation given in the output coordinate system. This shows how much the project was shifted based on the position of the GCPs and not how accurate is the reconstruction. 

Best,

my RMS error of absolute geolocation variance is increasing after adding DGPS GCP’s to the project, how can i minimize my AGV RMS.Also one of my friend says its increasing because gcp’s are correcting the incorrect positions/geolocations with respect to their initial positions thatswhy variance/offset is coming but my understanding is GCP are correctig the data nearer to it. All other data away from GCP is not changing much and causing more errors of geolocation variance. please eloborate on the issue.