GCP errors

[marcusrjc2]

Hi,

Can anyone help me with GCP errors

control points
Label    X error (m)    Y error (m)    Z error (m)    Error (m) Projections Error (pix)
C1        0.000010       0.000732       -0.000071    0.000735      11              0.051016
L4       -0.000067      -0.000482        0.000043    0.000488       8               0.159664
R3       -0.001093     -0.000320        0.000017    0.001139       7                0.043866
R4       0.001150       0.000070        0.000011    0.001152      10               0.073578
Total   0.000794       0.000468        0.000043    0.000922      36               0.091314

check points
C2    -0.000528        0.002097       -0.002893    0.003612      6                 0.031889
L1    -0.000135        -0.001518       0.000142    0.001530      8                  0.152357
L2    0.001001         0.000195        0.001156    0.001542     11                 0.051527
L3    0.000463        0.000482        -0.001963    0.002074      6                 0.073814
R1    0.002123       -0.000832        0.003155    0.003892     10                0.133128
R2    0.002369       -0.002044        0.003837    0.004951     13                0.091439
Total 0.001392       0.001405        0.002525    0.003208     54                0.099965

I am not sure how the totals are generated?

Can I assume the errors are the difference between the given coordinates and the 3d model calculated coordinates for the control points?

Then once these coordinates are assigned to the control points (does a transformation take place) are the check point coordinates compared to the new model coordinates?

I am analysing changing lighting arrays and comparing to a laser scan with interesting results. This is a double check as I cannot assume the laser scan is perfect so I wish to check with total station coorduinates for the GCPs - again I cant assume these are perfect.

Any help much appreciated.

[Alexey Pasumansky]

Hello marcusrjc2,

Errors for individual points are calculated as a difference between Source and Estimated values.

Total error is RMS error for all the points in the set (calculated individually for control and check points).

The difference between the control and check points is in fact that control points are used for referencing/optimization procedures and check points aren't.

[Napo]

HI,

I am in the same problem dealing with errors and I am wondering how AGISOFT calculate XY combined error and Z, as well as the total error reported in the AGISOFT report table?

I could not find any documentation online, Can you shed some light on this matter??

Thanks,
Napo

[BobvdMeij]

HI,

I am in the same problem dealing with errors and I am wondering how AGISOFT calculate XY combined error and Z, as well as the total error reported in the AGISOFT report table?

I could not find any documentation online, Can you shed some light on this matter??

Thanks,
Napo

Napo,

Did you have a look in the user manual? It actually does write something on the meaning of errors in the Reference pane, on pages 35 and 36 for Camera Locations and Ground Control Points respectively. Please see the attached screendump.

The total error is indeed the Root Mean Square Error (RMSE) of the X, Y and Z error for each GCP combined. Although the formula is not given, it IS given for Camera Locations (see the marked texted in the screendump) and it is the same for ground control points.

How these individual errors are calculated is indeed a different story and much is still unclear. The software probably considers how much the position of each (manually) marked GCP pixel is altered after application of camera optimization. The question remains if Photoscan considers displacement for all images on which you have marked the GCP, or only the image that is eventually used to 'build' the orthomosaic.

We have been benchmarking Photoscan recently and found that the errors provided by Photoscan are rather accurate in comparison to manual and independent validation of the Check and Control points using Qgis and AutoCAD. That is, relatively. In an absolute sense the results are slightly off, but the errors given by Photoscan are generally marginally larger compared to independent validation using other software packages.