Gaps in exported DEM

[rob_s]

Hello,

I am attempting to export a DEM created from 36 aerial photographs. The 3D model in photoscan looks very good and the DEM looks mostly good when viewed in ArcMap and ArcScene. The problem is that part of the area in the model didn't output in the DEM and I'm not sure why. There are some areas of water which have caused disturbance but these have been filled - there is a large section which looks fine in Photoscan and is not water, but has not exported in the DEM.

Any help would be much appreciated!

Thanks,

Rob

[Kjellis85]

If you look at the pointcloud in photoscan, is there any areas that aren't covered? If this is so, then that is the reason. Photoscan will not create DEM where there is no spatial information, even if the holes are filled when meshing. But it will make texture on areas that are filled by meshing. Why this is I don't know, I have only experienced this myself and figured that it must be the case. If anyone knows any better explanation please let me know.

[rob_s]

Thanks for the reply, the area with no data in the DEM is covered by plenty of points and I don't think there are enough gaps to stop it from calculating the height. The model in Photoscan also seems to have reconstructed the height in this area quite accurately, it is just on the DEM export that the problem occurs.

[Alexey Pasumansky]

Hello Rob,

Could you please specify what export format are you using?
Exporting to GeoTIFF looks fine. But some areas have negative values (colored blue on the image attached), could you also make sure that the views you are using shows points with negative heights.

[rob_s]

Thank you - I see what I have done wrong now. Classic case of operator error!

I don't have the XY co-ordinates for my GCPs, only the elevation data. I input 4 GCPs based on graticules from a printed map, and assigned only X and Y values (I don't know the elevation of these parts). I was therefore telling Photoscan to produce a model based on 4 points with an elevation of 0 (ignoring all the GCPs with elevation but no XY values), which is why there were some negative values. This may also be why there were some gaps initially - if they were too far out perhaps they could not be represented in a DEM.

I have got around this by using the 4 XY GCPs to estimate the X and Y co-ordinates of the elevation GCPs, whilst retaining the original elevation details. I am now running the model based on the estimated XY values and the known elevation values (and not the 4 XY GCPs). I think this will work.

[rob_s]

I have run the model again, this time with the correct GCPs (X,Y and Z values). The resulting DEM looks the same as my original one - with the same gaps! This time the elevation values are realistic so it seems to be doing everything right, except cropping some parts of the final output. Any ideas?

Thanks.

[Alexey Pasumansky]

Hello Rob,

And does exported GeoTIFF DEM have negative values with the correct GCP coordinates?

Could you please also export Ground Control values and send them to support@agisoft.ru.

[rob_s]

There are no negative values in the GCPs and the DEM I produced is okay apart from the gaps.

Thanks.

[mwillis]

Did this problem ever get figured out?  I am experiencing the same thing with some of my UAV data.  This is the first dataset I've seen this problem in.  No GCPs in my dataset, just the onboard coordinates which appear fine.

I have a great mesh but when DEM export happens there are a handful of blank spots in the DEM.  These areas in the mesh have good geometry (not just filled holes). Strange

[Alexey Pasumansky]

Hello mwillis,

Could you please check if exporting DEM with unchecked "Crop invalid DEM" option still has the same problem.

Is there a good overlap for the whole area and are the holes in the DEM large or small?

[mwillis]

Alexey, after unchecking "Crop invalid DEM", the export works properly.  Never noticed that option but should have.  Thank you.  The holes were small.

I worked around it yesterday by adding some photos from higher flight paths with strong overlap.