Agisoft PhotoScanPro question

[Richard House]

Hi All,

 

I hope some folks that took part in the recent Photogrammetry workshop might see this and can help clarify a couple of questions I have with PhotoScanPro.

 

1- PhotoScanPro processing- Ground Control Settings> Is it recommended to change the tie point accuracy setting from the default of 4 to .1 immediately after loading the photos? If not, at what point?

 

2- Is there a correlation between assigning more tie points manually in more photos that results in a corresponding shorter processing time? This is something I seem to observe happening and was  wondering if this is a reason to mark every photo with a visible target before building a dense point cloud?

 

Thanks to anyone who can assist me with these.

 

Best,

Rich House

 

Senior Photographer

Yale University Art Gallery

richard.house@yale.edu

artgallery.yale.edu

 

 

 

 

 

[Taylor]

Hi Rich,

 

Regarding the first question, if my notes and understanding are correct, it was suggested that the tie-point accuracy in the Ground Control settings menu can be left at the default value of 4 pixels to start with, or reduced to 1 pixel.  This setting could be reduced as you proceed with iterations of Gradual Selection of points and Optimization.  Even if you're confident of the placement of your markers, Tom cautioned that reducing the tie-point accuracy too much (e.g., 0.1 pixel) initially would over-constrain the tie points and be counterproductive.  The objective is to allow iterations of Gradual Selection and Optimization to remove the points with the most error and to improve the reprojection error of the remaining points.  Maybe Tom, Mark, Carla, or other class participants can correct me if I'm off-target here.

 

Tom also suggested using the reconstruction error setting in the Gradual Selection box only once, early in the optimization process, to remove the weakest points which have poor base:distance ratios because the camera positions are close together.

 

Re:  your second question, I haven't made comparisons by setting markers first, but my routine has been to do an initial alignment using Generic pair pre-selection in the Align Photos dialog box.  Disabling pair pre-selection can produce better alignment initially, but can take a very long time--it took about 30 hours on my Mini (quad i7) for 608 photos (and then crashed!) vs. about 5 hours using Generic pair preselection.  After I let the software attempt the alignment first, I check the Tools>Camera Calibration dialog box to see if Photoscan grouped the images appropriately and place them into the correct calibration groups if necessary.  Then I select and realign images that are clearly out of place or which failed to align initially (indicated by 'NA' in the Photos pane).  

 

I add markers if there are images that still aren't aligned correctly after these steps.  If you have a large number of images, manually setting markers accurately can take a very long time and this will offset the reduced processing time.  Sometimes it's still necessary to set markers for project-specific reasons, as I had to do in 608 image for my project. 

 

There can be an advantage to letting Photoscan attempt the alignment initially without the aid of markers, because it can be a huge aid in setting markers.  If you set markers in any pair of aligned images, Photoscan will project those markers onto all the other aligned images in which they appear.  You can then filter images by marker and adjust the marker positions more easily.  As you click on each new image with the same marker, Photoscan automatically centers the marker at the same zoom setting in the images, so you don't need to go hunting for the marker positions on every photo.

 

I enjoyed taking the class with you and hope this helps,

 

Taylor

[Richard House]

Hi Taylor,

 

Thanks for your response. I have been lucky in finding that the software has been able to align all of my photos without manually adding markers to assist. I'm still working out some kinks in my understanding of the processing details but I had thought that the "change the Ground Control setting for tie point accuracy from 4 to 1" step was supposed to be "4 to .1"- It's good to get that correction.

 

On to more testing...

 

Thanks again!

Rich