Capturing all sides

[Tony Buser]

Are there any tips for scanning all sides of objects in a scene? It
seems as though as I move around, I can pretty reliably get about
2/3rds of the sides of objects sitting on a table. However it quickly
gets difficult to get the backs of all the objects. Tracking failures
become more common and/or the back sides just never resolve. Since
I'm primarily interested in using ReconstructMe for 3D printing, it's
pretty important to get all 360 degrees of things.

I've had some success using a turntable, however since the resolution
on a kinect is rather low, most things you'd want to scan using
ReconstructMe are too big to put on a turntable. Although scanning
people while they sit and turn on an office chair works great!

[Tony Buser]

Here's an example of the problem: http://www.flickr.com/photos/tbuser/6923810639/

The back side of the gnome refused to resolve, even though for instance the surface of the box he's sitting on did resolve all around him.

[Christoph Heindl]

Are there any tips for scanning all sides of objects in a scene?  It
seems as though as I move around, I can pretty reliably get about
2/3rds of the sides of objects sitting on a table.  However it quickly
gets difficult to get the backs of all the objects.  Tracking failures
become more common and/or the back sides just never resolve.  Since
I'm primarily interested in using ReconstructMe for 3D printing, it's
pretty important to get all 360 degrees of things.

Are you moving or are you using a turntable like helper? In the second case, make sure that everything the camera sees turns, or you will get messed up results. Another idea is to add something complex to the scene (clothing) to stabilize the tracking. I didn't mention on the docs that pure spherical and pure cylindrical (in essence symmetric objects) will cause the tracking to fail. You need to make sure that not only symmetries are visible during scanning. Often that means more postprocessing work to remove the background, but we could use this technique in the future to automate things

http://reconstructme.net/2012/02/11/3d-segmentation-using-reconstructme/ 

It is on our todo list, but won't be added in the near future (one month)

 

I've had some success using a turntable, however since the resolution
on a kinect is rather low, most things you'd want to scan using
ReconstructMe are too big to put on a turntable.  Although scanning
people while they sit and turn on an office chair works great!

Just make sure that everything in the sensor frame turns.

Best,

Christoph

[Charles Dunk]

Could part of the problem be that the gnome is reflective? I've found
reflective or transparent objects don't resolve well

[Tony Buser]

That gnome is made out of stone.  So I don't think that is it.  I think it's more likely because of the low resolution of the kinect, the front and back is too cylindrical and it doesn't pick up enough details to differentiate.

[Christoph Heindl]

I'd love to test this gnome :) I will look for a similar object in our office next week.

Best,

Christoph

[Tony Buser]

Here is what the gnome looks like by the way. I find gnomes to be
great test subjects for testing different 3D scanning systems. :)

http://www.flickr.com/photos/tbuser/6147457473/

On Feb 26, 2:11 am, Christoph Heindl <christoph.hei...@gmail.com>
wrote: