Forum

Author Topic: Most predicable set for better alignments  (Read 6803 times)

pyruss

  • Newbie
  • *
  • Posts: 6
    • View Profile
Most predicable set for better alignments
« on: September 03, 2014, 12:43:06 PM »
Hi guys i have a 109 camera system for body scan
And i was testing agisoft?s  software

What would be the best set for the best possible camera alignments?

1- Masks on or off?
2- with or without markers?
3- pre-selection pair enable or disable?

I have made many tests but results arent conclusive and i was trying to find a working pipeline that i could use in a production enviroment

what is your experience?

Lambo

  • Full Member
  • ***
  • Posts: 167
    • View Profile
Re: Most predicable set for better alignments
« Reply #1 on: September 05, 2014, 08:37:11 AM »
Hi Pyruss, I normally use pair pre-selection OFF, no markers ( I don think they are needed for body scans) and using Masks increases the detail and quality of the results in my case.
Now, if the scan was not too good because the clothes didnt read well, the person was too small so you didnt get enough overlap in some areas, etc. I found that no Mask helps it get the correct alignment easier since it still has the background to use for the calculation.
Just my 2 cents.
Leo

Digitage

  • Newbie
  • *
  • Posts: 10
    • View Profile
Re: Most predicable set for better alignments
« Reply #2 on: September 09, 2014, 08:28:11 PM »
Hi Pyruss,

I concur with Lambo, masks can be useful when the subject is not detailed enough to give you a perfect alignment, but can be swithed off otherwise.

You might want to clean up the point cloud and use the optimise alignment process systematiocally as well after a first alignment calculation. It can improve results tremendously.

Hope this helps.

FLuca

  • Jr. Member
  • **
  • Posts: 55
    • View Profile
    • Another-me
Re: Most predicable set for better alignments
« Reply #3 on: September 10, 2014, 11:59:18 AM »
Hi Digitage,
I use PS for many months now with acceptable results but I think I have missed something.
How do you optimise the cam alignment ?
Is it a feature of the pro version?
In the standard version I have not found such feature.
Do you mean selecting cameras and re-applying 'Align Selected cameras' ?
I do this systematically and apparently it does move the cam slightly so I suggest they are optimised. But with this procedure, all points removed manually from the sparse cloud seems to be used in the alignement as they appear again in the new spare cloud. Therefore I do not see the point in removing points with gradual selection.
Could you please detail a bit more your optimisation step ?
Thanks
Fred Lucazeau - Body Scan - www.another-me.fr
Sketchfab gallery: https://sketchfab.com/fredlucazeau

FoodMan

  • Sr. Member
  • ****
  • Posts: 477
    • View Profile
Re: Most predicable set for better alignments
« Reply #4 on: September 10, 2014, 01:56:22 PM »

How do you optimise the cam alignment ?
Is it a feature of the pro version?


yes pro version only..

f/

FLuca

  • Jr. Member
  • **
  • Posts: 55
    • View Profile
    • Another-me
Re: Most predicable set for better alignments
« Reply #5 on: September 10, 2014, 03:47:25 PM »
Too bad !
So what is the purpose of gradual selection for the standard version ?
Fred Lucazeau - Body Scan - www.another-me.fr
Sketchfab gallery: https://sketchfab.com/fredlucazeau

Lambo

  • Full Member
  • ***
  • Posts: 167
    • View Profile
Re: Most predicable set for better alignments
« Reply #6 on: September 11, 2014, 02:51:21 AM »
Gradual selection helps the dense cloud to be a little cleaner. It removes points that are not totally accurate that can create some noise in the resulting dense cloud and mesh after that.
When I have a very good scans with enough points on the sparse, I always do the optimize since the resulting mesh comes out nicer. BUT if the scan was not too good and I don't have enough points, then I do not do the optimize since it seems to come out missing some areas after that.
Leo

meshmixup

  • Jr. Member
  • **
  • Posts: 78
    • View Profile
Re: Most predicable set for better alignments
« Reply #7 on: September 11, 2014, 01:06:37 PM »
 I like to keep most dense points and just manually remove those shells of points which are obviously offset from the main cloud. I do not use gradual selection becos I will do the cleaning later anyway ;)
« Last Edit: September 11, 2014, 01:16:44 PM by meshmixup »

Olga

  • Newbie
  • *
  • Posts: 1
    • View Profile
    • ScanLab Photogrammetry
Re: Most predicable set for better alignments
« Reply #8 on: September 14, 2014, 01:36:25 AM »
Hi guys i have a 109 camera system for body scan
And i was testing agisoft?s  software

What would be the best set for the best possible camera alignments?

1- Masks on or off?
2- with or without markers?
3- pre-selection pair enable or disable?

I have made many tests but results arent conclusive and i was trying to find a working pipeline that i could use in a production enviroment

what is your experience?

Hi pyruss,

Here are some suggestions:

1. It all depends on the equipment you are using, focus and clarity of your images. If your picture is of a low resolution, there is not need in using masks. Allow photoscan to take full use all the information present in every image. If you think that your cameras have enough resolution, then go ahead and mask, the results are going to be better.

3. We always set pre-selection to ON to allow Photoscan to use downsampled version of the image to save on the computation time. 

Good luck,
Olga
Olga Kanapelka
ScanLab Photogrammetry | olga@scanlab.ca | http://www.scanlab.ca

pyruss

  • Newbie
  • *
  • Posts: 6
    • View Profile
Re: Most predicable set for better alignments
« Reply #9 on: September 16, 2014, 03:53:23 AM »
Hi Guys

Thanks for all the Feedback

I am using all Nikon D3200 + 18-55mm lens (going to change this for the 50mm prime)

The images do lack sharp but are reasonable

My current pipeline is

0- shoot in Raw (NEF) high quality
0a- use camera raw to push some sharp to all images and convert to TIF

1- Inicial alignment with pair pre-select
2- create model from sparse cloud
3- mask photos against low res model
4- look for un-aligned photos
5-try to re-align those photos (individualy)
6- erase photos that cant be aligned
7- re-align (full) with mask on and pair pre-select off
8- create dense cloud (high)
9- build mesh (high)
10- if mesh lacks detail
10a - re-create masks from more acurate model
10b- re-create dense could (high)
10c- build mesh


p.s- i use markes on the floor to help with scaling