convex DEM and rifts in DEM

[ferdywallinx]

I am using Photoscan to build DEM's from aerial imagery and at moments i get the following problems:

The DEM that photoscan produces is convex (see attachment) or tilted.

The DEM shows rifts (height can drop over 60 meters within one meter.

Did anyone else experience these issues and what is your solution? Note: ground control points are not possible for me.

[stihl]

There could be multitude of reasons for this happening. Such as misalignment of the photos because of too little texture, too much or too little contrast, photos too bright or too dark, rolling shutter effect, aperture varrying wildly, motion blur, use of autofocus or auto stabilisation, wrong photo positions, wrong coordinate system, photo position columns mixed up, incorrect camera calibration parameters..etcetcetc

Did you use photo positions before aligning? What resolution are the photos and what's the overlap? Producing a DEM without GCPs or photo positions is rather pointless as it has no real value.

[ferdywallinx]

Dear Stihl,

thank you for your reply. The images are good in that they have no blur, no oversaturation or strong noise, no rolling shutter (D810 DSLR camera), fixed aperture and no autofocus or autostabilisation. The coordinate system and the photo positions are also correct (EPSG:4326 format). The overlap in the area with the rift is 7-9. We have not calibrated the camera's and the GPS shows strong inaccuracy in the photoscan report in the southwest area but this is also where the rift is occurring. Therefore i am wondering why photoscan pulls the images in the west down (accuracy of -47 meters) and up in the east part of the southwest area the gps position is pulled up and i can imagine that this causes this rift (see attachment).

Is there any way that i can prevent photoscan from doing this and if so should I?

[stihl]

That's interesting. Which version of Photoscan are you on?

Are the photo positions all taken at roughly the same altitude or are you seeing a difference in altitude coordinates on the location where it gets pulled down? Are you seeing this change in altitude in the projected photos by turning on the option "Show cameras"?
Can you share a screenshot of the tie-points and the photo positions and perhaps a close up of the tie-points where this rift occurs?

[Alexey Pasumansky]

Hello ferdywallinx,

Is the information about camera orientation angles input to the Reference pane?

[ferdywallinx]

Hi Stihl and Alexey,

unfortunately i just broke the project (don't know how, photoscan keeps crashing), so i cannot share any more screenshots (running the project takes 2 days). no orientation data was loaded but nothing strange was to be seen on the sparse point cloud. i have attached overlaid the location of the cameras in altitude. though you can see that the pilot went down one way and up the other way (flightline) i do not know why this would create such big gaps.

[Alexey Pasumansky]

Hello ferdywallinx,

Please submit the crash report and include your e-mail or forum nickname, so that we can track it easily.

[Alexey Pasumansky]

Hello Hello ferdywallinx,,

Are you using remote desktop for connection? If so, then I can suggest to switch the default view mode to Point Cloud before loading any project, or consider using other remote connection options (like Team Viewer, for example). Also I suggest to update to the version 1.2.5 build 2735.

[ferdywallinx]

Hi Alexey,

I did all these things and the project is loading. This does still not solve any issues regarding rifts and convex dems so any help is still very much wanted.

[Alexey Pasumansky]

Hello ferdywallinx,

Then you can maybe send the report file and copy of the project saved in PSZ format with only alignment results (without dense cloud and mesh) to support@agisoft.com?

[Alexey Pasumansky]

Hello ferdywallinx,

Sorry for the delay.

We have checked the project and assume that the issue is related to the incorrect coordinates input for the camera locations. Also it may be caused by the synchronization issue between the camera shutter and GPS device or considerable distance between them, if the common aircraft is used. In the report file there's quite obvious indication, that the planar errors for the camera positions are pointing into different directions.