Author Topic: Gradual selection - need a simple definition  (Read 5245 times)


  • Jr. Member
  • **
  • Posts: 74
    • View Profile
Gradual selection - need a simple definition
« on: April 27, 2013, 07:30:09 AM »
Hello all,

I am struggling to understand the principles behind the 2 options available within Gradual Selection, specifically in point cloud mode.

I have read what manual has to say about it but still don't fully understand.  I process aerial imagery of areas up to 50ha from a UAV, mines and quarries mainly, and believe that this process is going to be important to me in achieving the best results.

Can someone possibly explain these features in 'very basic' language to me and perhaps even suggest what settings I should follow to remove erroneous points etc? 

What exactly do the numbers on the scale bars mean when performing both Reprojection Error and Reconstruction Uncertainty?  Reprojection Error I am guessing is a distance in metres?

Any explanation would be much appreciated!

Many thanks,


  • Newbie
  • *
  • Posts: 13
    • View Profile
statistical foundation
« Reply #1 on: May 14, 2014, 01:31:08 AM »
Working on my masters thesis and wanted to go into/understand the statistical foundation for point cloud refinement a bit better.  Specifically, what's the bases for point selection when using "reconstruction uncertainty" and "reprojection error".  Any help or links to papers (published or not) would be great.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14615
    • View Profile
Re: statistical foundation
« Reply #2 on: May 14, 2014, 07:23:59 AM »
Hello Aaron,

Reprojection error
High reprojection error usually indicates poor localization accuracy of the corresponding point
projections at the point matching step. It is also typical for false matches. Removing such points can
improve accuracy of the subsequent optimization step.
Reconstruction uncertainty
High reconstruction uncertainty is typical for points, reconstructed from nearby photos with small
baseline. Such points can noticably deviate from the object surface, introducing noise in the point
cloud. While removal of such points should not affect the accuracy of optimization, it may be useful
to remove them before building geometry in Point Cloud mode or for better visual appearence of the
point cloud.
Best regards,
Alexey Pasumansky,
Agisoft LLC


  • Newbie
  • *
  • Posts: 13
    • View Profile
Re: Gradual selection - need a simple definition
« Reply #3 on: May 14, 2014, 10:44:30 PM »
Hey Alexey,
That helps clarify my definitions a bit but doesn't quite answer my question still, I don't think.  I also didn't see anything in the conversation you linked.
As it stands, I've had great success working with both gradual selection tools, but it's still a bit of a "black box" to me.  I understand that it refines accuracy/point placement, but I am interested in how.  Essentially, how does the software "know" that a point is poorly localized and that deleting it will make the optimization (and therefore accuracy) better?