working with pattern projection

[thomas83]

i am currently working with pattern projection and stereofotografy to aid photoscan in the process of finding points when working with homogenic surfaces.
i use a led beamer to project a random noise pattern on a surace, then i take a stereopicture.
i use different chunks for each projection/image pair and merge them. this whole chunkmerging is really tiresome and i have to use manual markers, which makes the whole effort hardly worth it.

is there a way to tell agisoft to use these stereo image pairs instead of trying to find corresponding points across all fotos? maybe via the file name, like cam1_01.jpg and cam2_01.jpg? so that i can work on one chunk i mean...  i have seen this option in programs like 3DSOM.

also, would it be possible to use one set of photos to find the points/build geometry and a second set to build the texture?

thanks again!
tom

[Wishgranter]

as far as i tested the PROFESIONAL version, you can set coresponding points, try demo and it should work, its like local coordinates....... have tested few weeks ago, it but no time now for deeper retesting...... if something just PM me.....   

[thomas83]

thanks for the quick reply wishgranter, but thats not what i meant...
i am trying to avoid setting markers by hand. basicly what i?d like to tell photoscan to do is this:

a) look for corresponding points between foto 01 and 02 (i.e. one stereopair)
b) look for corresponding points between foto 03 and 04 (i.e. next stereopair)
...etc...
c) allign and merge all found points (without having to set markers and before building geometry)
d) build geometry

i?m trying to project the pattern and take a stereopair-picture, and then moving on to the next angle.
as of now i have to create one chunk per angle so that photoscan only looks for corresponding points in the stereopair.... is there any way to to tell photoscan which fotos to pair up when looking for points?

[Alexey Pasumansky]

Hello Tom,

Could you please specify if the projected pattern doesn't move relatively to the object on the different stereopair images taken?
In case the projection is different each time the automatic aligning of stereopairs (chunks) is not possible. Maybe in this case it is somehow possible to fix the projector so the pattern on the whole object doesn't change during the shooting process.

[thomas83]

hello alexey,

the projected pattern is different for each stereopair, since i'm using a turntable with white backdrop to turn the object and a fixed camera/projector... so there is no possibility for influencing the pair selection? if not i guess i'll have to manualy set markers and merge the chunks then

would it be possible to swap pictures after positions and points have been calculated? heres an example:
1) taking a stereopair with projected pattern
2) taking a stereopair from the same position without pattern
3) align stereopair (with pattern)
4) swap pattern-stereopair with "clean"-stereopair for texturing

[Alexey Pasumansky]

Hello Tom,

More likely that PhotoScan would not be able to align photos of the same object but with different texture.

And as for the picture swapping it is possible to change photos with the projected texture by "clean" images in the folder they are kept. It should be done when the geometry is already reconstructed just before the build texture step. And it is necessary that each pair of clean/textured photo was taken from the same position and have absolutely same filenames.

[Wishgranter]

Thomas, try the demo of Professional version, set some markers that are clearly visible on photos, no too much, 4-6 on one photo clearly to recognize, square or cirlces. then try to connect them on photos, correct if some are not precise and procces the scene, it should help to recognize positions of camera and help in the process of reconstruction.
 
Shiny homogeneous surfaces are big problems, so you must use ways to go around it. another possible solution is to use self adhesive stamps, something like a chess board like or the ones in Calibration - http://www.agisoft.ru/products/lens/ that are clearly visible - consider in what resolution you shoot. this is like you project patterns over Projector, but simpler to achieve.... start with smaller part of car or what are you shoot, learn from that, and use on full model. Let us know if worked.

You should read a few principial things about Photogrametry so you can understand basic principes of the process, and overcome limitations - sometimes :-)

Here is a few pages about what is need to archieve, but understand this is another way, another proces of working with data, this process use everytime different pattern you MUST use the same on all photos.. just to have idea about how big-small should be patterns..

http://en.wikipedia.org/wiki/Structured-light_3D_scanner
http://www.cadtechsolutions.in/3d_whitelight_scanning
http://www.sciencedirect.com/science/article/pii/S0924271611000414

[kosmarnik]

If you're reconstructing smaller objects (as long as they fit in your studio), you'll get much better results with a structured-light approach, though shiny/reflective/light absorbant surfaces are a problem (though, you can coat them with a matte coating for perfect results).
SFM approach is better for buildings and large scale 3d reconstruction.

[thomas83]

yes you are right, i?m exploring that option too... what bothers me about the sl-approach is the bad texture resolution you get when using a webcam (e.g. DAVID-Laserscanner), high priced ccd-cams are not an option for me... not to mention the dimensional limitations of a calibration corner.. i am trying to automate the process with an arduino-controlled stepper and shutter release for sl-scanning with my fuji finepix hs20, taking a photo for each projected pattern... nevertheless this is very time consuming...

[Joules]

We recently did a motorbike fuel tank that had been primed, then used a black white board marker to draw a grid on the tank, worked very well for us.  We have played with structured light, but it was just a No No in regard setting up, working round an object and keeping everything in sync.  Total ball ache... (technical term for spherical digitising) 

[kosmarnik]

yes you are right, i?m exploring that option too... what bothers me about the sl-approach is the bad texture resolution you get when using a webcam (e.g. DAVID-Laserscanner), high priced ccd-cams are not an option for me... not to mention the dimensional limitations of a calibration corner.. i am trying to automate the process with an arduino-controlled stepper and shutter release for sl-scanning with my fuji finepix hs20, taking a photo for each projected pattern... nevertheless this is very time consuming...

Yes, DSLR is the way to go with SL. Much better dynamic range and textures.
WRT calibration corners, ppl have used very large ones, but your scan res is limited to the projector resolution. Laser scanning would get you better res but alas can't be used with a DSLR