Forum

Author Topic: [solved] Large "faces" with height field DEM processing  (Read 14386 times)

nanders

  • Newbie
  • *
  • Posts: 25
    • View Profile
[solved] Large "faces" with height field DEM processing
« on: March 13, 2013, 07:20:26 PM »
Hi all,

I just started using Agisoft Photoscan Pro (0.9.0) for processing UAV data for creating digital surface models with a 6-core XEON processor and 32GB RAM. I am pretty impressed by Photoscan's ability on automating this workflow, but run into a few problems.

According to the documentation the algorithm "height field" is optimized for creating terrain geometry.
A typical UAV flight contains 1000 jpeg images of 16MP, the largest area consists of 4 sequential flights with app. 4000 images. The height field algorithm is indeed fast and low on memory. However, the result is not satisfying as it seems the point cloud is imported into a TIN and then simply converted to a digital surface model (distinct faces are present in the terrain model). The "arbitrary" method gives far better results (see figure), but 32GB of RAM is not enough to even process 800 images (photoscan is killed with 'out-of-memory', 400 images seem to work fine).

Is there a better way to process my 4000-images-project, other then using arbitrary mode and split my area into 10 subprojects? Or is the memory management of arbitrary mode better in v0.9.1? As all the images are in one folder (ordered in flight lines) it is quite some work to localize all images and create subfolders for creating chunks.


The figure shows a shaded reilief image of the digital surface model that represents a landslide on a hillslope.

« Last Edit: March 14, 2013, 09:41:44 AM by nanders »

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 15276
    • View Profile
Re: Interpolation techniques for DEM processing
« Reply #1 on: March 13, 2013, 07:23:55 PM »
Hello nanders,

Could you please specify the face number parameter used for Height Field reconstruction?
Seem like the value is very low. We recommend to use 1 - 10 Millions of faces for such dataset.
Best regards,
Alexey Pasumansky,
Agisoft LLC

nanders

  • Newbie
  • *
  • Posts: 25
    • View Profile
Re: Interpolation techniques for DEM processing
« Reply #2 on: March 13, 2013, 07:33:51 PM »
thanks for the quick reply. I used 0 (I thought that meant unlimited because I have no clue on what my max should be).

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 15276
    • View Profile
Re: Interpolation techniques for DEM processing
« Reply #3 on: March 13, 2013, 07:43:05 PM »
Hello nanders,

in the version 0.9.0 the meaning of "zero" face count value was changed for Height Field mode, now it refers to automatically calculated (default) value.
In this case the triangulation process continues until the difference between values in DEM grid and triangulated values are less then half of a grid step.
So if you require the maximum number of faces (similar to "zero" face count in 0.8.5) you need to input considerably greater values (for example, 100 Millions).

Best regards,
Alexey Pasumansky,
Agisoft LLC

nanders

  • Newbie
  • *
  • Posts: 25
    • View Profile
Re: Interpolation techniques for DEM processing
« Reply #4 on: March 13, 2013, 07:47:58 PM »
Thanks for the explanation! I'll try that and will post the result here..

nanders

  • Newbie
  • *
  • Posts: 25
    • View Profile
Re: Interpolation techniques for DEM processing
« Reply #5 on: March 13, 2013, 08:30:57 PM »
I now used 100M faces with the height field mode, and 0.5 filter value (is this the 0.5 grid cell threshold?), and it indeed looks a lot better.

Thanks!


RalfH

  • Sr. Member
  • ****
  • Posts: 344
    • View Profile
Re: [solved] Large "faces" with height field DEM processing
« Reply #6 on: March 14, 2013, 11:31:41 AM »
Nice landslide indeed. If you zoom out a bit, are the faint undulations below traces of older landslides?

nanders

  • Newbie
  • *
  • Posts: 25
    • View Profile
Re: [solved] Large "faces" with height field DEM processing
« Reply #7 on: March 14, 2013, 04:32:57 PM »
I did not immediately recognize this in the field, and these landslides are pretty shallow and small ones get plowed away every year, but yes, based on the surface model it looks like remnants of older ones. The UAV data shows lots of interesting features :-)