Vertical error in DSM to Ground Control

[kgrover]

We recently processed a site from a UAV, and placed a number of ground targets surveyed with RTK. We held 5 of the 10 points in Agisoft, and left the others as checks. The results from the processing in the Reference pane in Agisoft looked great (attached file Agisoft Errors). We created a DSM and exported as a Geotiff, then started doing further comparisons and noticed a vertical issue.

We determined the elevations of the 10 aerial targets from the exported DSM, and are finding around 5-8 cm in vertical differences from the DSM to our ground targets (attached file Error Check).

I would assume that our resulting DSM should give us close to identical errors as shown in the reference pane in Agisoft? Why would it be different?

Kevin

[kgrover]

I wanted to give this one a bump again, since I haven't heard from anyone.

Any ideas??

Kevin

[Alexey Pasumansky]

Hello Kevin,

What settings you've used to build dense cloud, mesh and which parameters were used in the Export DEM dialog?

[kgrover]

Hi Alexey,

I used the following settings:

Dense Point Cloud - Medium with moderate depth filtering ( also tried low and high and got same results)
Model - Used height field with dense cloud, with a high face count
Export DEM - exported using as given pizel size values of 0.12m

I tried this in the version 1.1.6 version, as well as the new beta release, and also changed a number of the parameters as above, and it all gave me similar results.

I also tried just exporting the Dense Point Cloud and create a surface in third party software, and it had the same issue. So it seems like it is happening directly from that step.

Kevin

[Alexey Pasumansky]

Hello Kevin,

Maybe you should use more markers with better distribution on the model surface, then optimize the camera alignment results prior to the dense cloud generation.

In the version 1.2 you'll be able to build DEM directly from the dense cloud, and it will help to avoid any issues related to the improper bounding box orientation (its' red colored side) and the large polygons, if the number of polygons is rather low.

[nadar]

Vertical reference in WGS84 system is sometimes based on an ellipsoid, and sometimes inegrates a local correction to better fit the geoid. In some regions, the difference can reach 50 m

[photoscan_user]

What is general procedure to compute errors of DSM regarding to geoid?

[AThorp]

Kevin,

What sort of control did you have on the UAV flight regarding determining the coordinates of the camera when the picture was taken?  Just plain GPS, or did you have RTK on the drone as well?

Al