error meaning

[ascreus]

Hi at all,

please help me in understanding the "error" meaning.

As you can see in picture attached, the point 4 altitude is different from that measured on DEM (lower right): dem is 5 cm higher than expected from target.
If I check the errors tab, I can see that is only 18mm..

so, How should I consider this reading?

thanks

[olihar]

Are you using the correct GEOid for the images?

[ascreus]

I think so...

I converted the location of each image starting from wgs 84 (as obtained from exif of Autel Evo 2 images) to EPSG_6707

the same for the targets

[Paulo]

Hi ascreus,

the DEM will never be as precise as the fit from your ticked GCPs...

Depending on your camera and procedure., the DEM error could be in the range of 1 to 3 GSD.

What is your average GSD? and what quality are your depth maps generated in?

Example of Surface validation (point cloud, model, dem) using 10 GCPS with flight at 5.6 cm resolution and high quality depth maps :

Code: [Select]

  AVG GSD (cm/pix): 5.587
 ----------------------------------------------
               Surface Validation             
 ----------------------------------------------
 Dense Cloud
 Point   Marker_Alt   Surface_Alt  Error_Alt 
 1       303.077      303.082      -0.005     
 3       302.376      302.401      -0.025     
 4       302.407      302.407      -0.0       
 5       300.954      300.987      -0.033     
 6       299.716      299.761      -0.045     
 7       298.768      298.81       -0.042     
 8       300.636      300.566      0.07       
 9       302.519      302.589      -0.07     
 10      302.517      302.578      -0.061     
 11n     299.93       300.002      -0.072     
 10 Pts   AVG: -0.029  STD: 0.04    RMS: 0.049
 ----------------------------------------------
 loaded mesh in 1.201 sec
 Model
 Point   Marker_Alt   Surface_Alt  Error_Alt 
 1       303.077      303.063      0.014     
 3       302.376      302.385      -0.009     
 4       302.407      302.407      0.0       
 5       300.954      301.005      -0.051     
 6       299.716      299.756      -0.04     
 7       298.768      298.807      -0.039     
 8       300.636      300.555      0.081     
 9       302.519      302.552      -0.033     
 10      302.517      302.532      -0.015     
 11n     299.93       299.989      -0.059     
 10 Pts   AVG: -0.015  STD: 0.039   RMS: 0.042
 ----------------------------------------------
 DEM
 Point   Marker_Alt   Surface_Alt  Error_Alt 
 1       303.077      303.046      0.031     
 3       302.376      302.4        -0.024     
 4       302.407      302.404      0.003     
 5       300.954      300.986      -0.032     
 6       299.716      299.743      -0.027     
 7       298.768      298.811      -0.043     
 8       300.636      300.564      0.072     
 9       302.519      302.587      -0.068     
 10      302.517      302.577      -0.06     
 11n     299.93       300.021      -0.091     
 10 Pts   AVG: -0.024  STD: 0.046   RMS: 0.052
 ----------------------------------------------RMS errors are consistently below 1 GSD...

[ascreus]

Do you think I need to try with higher quality?

in any case... and which error is considered in the table?

thank you

[ascreus]

Nothing to do.

Also in "high" mode, the resulted dem is locally up do 20cm higher than GCP.
The problem is that the area is quite big (30Ha) and 20cm could result in a big amount of cubic meter!!

(I need to compare two dem for the same region after one month, where only few part are changed...)

Please help

Antonio

[Paulo]

Hello Antonio,

i would check or validate your DEM against all GCPs not just one...Is the error of 20 cm consistent or just a few?

If error is only on a certain area then you can maybe better grasp the issue...

[ascreus]

Hi Paul

resolution of dem is 9.88 cm/px

please see table and image attached

maybe could be fine....
but when I compare this dem with the january (one month before) one, the differenze resoult in up to 25 cm also where no change has occoured
maybe the "sum" of dems tolerance in respect of gcp?

antonio

[Paulo]

Antonio,

yes certainly a consistent offset between 2 surveys...

May be if you can export the markers reference as tab separated text including Errors, Accuracy, Estimated and Enabled for both surveys and share. Example of Exported reference file attached...

[ascreus]

here we are

[Paulo]

Antonio,

definitely the  february GCPs Z do not agree with the estimated accuracy (default) of 0.005m. If they were determined at sub cm accuracy they should have Z errors  in the cm range and in your case they are in 5 cm range:
Count: 8
Minimum Value: -0.030063
Maximum Value: 0.069252
Average Value: 0.020518625
Standard Deviation: 0.032863743
Sum: 0.164149

First I would set the accuracy of your GCPs to something more realistic as 0.005 m is too optimistic (maybe 0.01/0.02) and then update transform to see your results...Also untick one of your GCPs (set as check point) and update and see the error you get on this point..

In January the Z errors are much tighter and reflect something consistent

Count: 7
Minimum Value: -0.002993
Maximum Value: 0.001796
Average Value: 0
Standard Deviation: 0.0015116766
Sum: 0

[ascreus]

Hi

please have a look at error file after introducing more gcp (from stable zone revealed in january survey) and after set 0.01 value for accuracy

although a general improvement is obtained, locally difference between gcp and dem is up to 21cm....


I do not understand

[Paulo]

Ciao Antonio,

credo che é un problema de levantamento.... En la zona de pontos T7 y 8, los errors de Z son muy grandes....

Also I am wondering how these points were surveyed? AS there are also high errors in X or Y in many T points....

Please contact me with Personal Message for more info...

Maybe if you send me your Metashape processing reports for january e febrero I can possiblemente capire piu!

Excusa meu italiano e piu spagnolo che altra cosa!

[ascreus]

Hi Paulo

I'm writing here in order to share the solution with those interested in.

During the drone survey I lost the rtk signal for a few minutes; this caused a drop in accuracy for a numbers of images. Suddendly the drone recovered the rtk correction before the survey ended.

Please have a look at the image below..

I didn't consider it to be a big problem...hoping for a compensatory effect of the entire set of images..
but I was wrong!
Evidently the low accuracy images affected all the survey, generating big differences also in photo taken quite distant


By removing the last three stripes, I obtained the desired quality, reducing the difference in "no change" zone to a 2-3 cm!

Ok...
There is ever a human error :-)


Thank you very much

And your "esperanto" was very understandable 

[ascreus]

but... in any case...
the question is:

when I read a value in the error column, what is the real meaning?
why I obtain a significative difference in between this value and the gcp to dem value?

thanks