Forum

Author Topic: How many points per image should I detect?  (Read 5624 times)

airmap3d

  • Jr. Member
  • **
  • Posts: 74
    • View Profile
How many points per image should I detect?
« on: January 19, 2013, 12:58:09 PM »
Hi all,

Can anyone help me with determining how many points per image that I should detect?  I see the default is 40,000.  I fly low level UAV aerial photogrammetry with a 20mm lens and images around 8.5Mp each.  I can only assume that less points will mean faster processing but less accurate orthos and DEM.  Is 40,000 the recommended or should it be higher or can I get away with less?  Obviously the faster I can process my data the better but I do not want to compromise too much on output quality.

Thanks!

RalfH

  • Sr. Member
  • ****
  • Posts: 344
    • View Profile
Re: How many points per image should I detect?
« Reply #1 on: January 19, 2013, 03:13:51 PM »
The points per fotos refer only to the alignment step, so they don't have so much relevance for the final model or DEM. And yes, using less points per foto will result in faster processing, in particular in the matching step. In the model building step, this should not really have an influence on processing time and model quality. I always leave it at 40,000, but for large sets of images (many hundred or even thousands) it may be reasonable to reduce the number a bit. If you go too low, I suspect that the risk of false matches and resulting alignment (and perhaps camera calibration) errors would go up. I assume that the Agisoft people have done some testing which led to the preset number.
« Last Edit: January 19, 2013, 03:39:07 PM by RalfH »

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: How many points per image should I detect?
« Reply #2 on: January 19, 2013, 03:50:28 PM »
Airmap, as many points then better.But have read some whitepaper that more than 1000 is waste of resources, BUT, it seems that Pscan create something as group of interest, say afther points are on more than say 10 images then pscan create groups from that, its good because it can then align photos that are hard to do in other sw. its just my deduction afther have used photos from air 200m + and ground photos....  AGISOFT team can ecentuely explain better... if they can..
----------------
www.mhb.sk

airmap3d

  • Jr. Member
  • **
  • Posts: 74
    • View Profile
Re: How many points per image should I detect?
« Reply #3 on: January 20, 2013, 05:54:08 AM »
Hi guys, thanks for the replies.  That was what I suspected.  I might just do some testing myself and see what sort of differences it makes having less points.  I guess the best way to know something for sure is to test it yourself!  Maybe if the photoscan team read this post they will be able to give me more 'inside information' as to where the 40,000 came from.

Cheers.