transparent objects for ant experiment

[feinermo]

Hello,

I am a scientist studying ants. I am now starting to study weaver antnest building. These ants construct their nests by placing leaves next to each other and weaving them together to make a sphere-like objects. To do this I would first like to acquire a 3D time-lapse visualization of the construction process.

The main problem is that as the leaves are placed within a structure they naturally obstruct each other from the camera view. To resolve this i would like to use semi-tranparent artificial leaves. This would allow me, at least in theory, to visualize the entire 3D structure of the ants' nest.
I understand that doing photogrammetry on transparent objects is far from being straightforward however there are several things which may work in my favor.

- i intend to use a multi-camera system with fixed cameras around the objects. The cameras may be calibrated on a simpler opaque object before the actual measurement. (this should also solve the problem of filming a moving object).
- the leaves are semi-transparent so that they have a color - i can use different colors for different leaves
- another possibility is to use full transparent material for the leaves but draw a grid on each leaf. again, i can use different color grids for different leaves.

I hope my description is clear enough and would appreciate any advice.

thanks!

[Andrea Alessi]

Hi feinermo,

I'm not sure I understand, sorry! do you want to reconstruct the surface of the leaves themselves or what lies behind them?

If you want to reconstruct the leaves, you may get away with semitransparent ones. I would avoid transparent leaves - grids are likely not going to picked up unless you make the very textured (e.g. perlin noise) and even if they do, you still won't be able to reconstruct the surface of what is transparent.

If instead you are trying the reconstruct a structure built behind a leaf, then you should use a transparent one.

Regardless, it's definitely going to be challenging - if you can give a few more details I'd be happy to help.

[feinermo]

Thank you Andrea for your quick answer!

Sorry for not being clear enough: Yes, i do mean to reconstruct the leaves themselves.
I attached two photos to make it easier to understand what these semi-transparent leaves might look like at end of the process (i can also try other materials the ants will work with any flexible surface).
You can see a outline around each leaf - this was added by a black marker and, of course, i could add different patterns as required.
Further, as in my original post, i have no problem to use multiple fixed cameras positioned all around the "plant".

The 3D location of the ants themselves is, of course, also interesting but as a first step i would be happy to just get a reconstruction of the leaves.

If you advise - i can certainly try taking pictures of these semi-transparent multi-leaf objects and feeding them into the software just to see what happens.

thanks!

[Andrea Alessi]

Hi feinermo,

thank you for your additional information!

I think what you want is to have the structure *of each* leaf, including those occluded. This is going to be very hard if not downright impossible.

While you might get some structure, it won't be a full, smooth reconstruction of the external leaves if you use those semi transparent ones (you can get very good results with real leaves). However, those underneath are not going to be reconstructed.

However, what you may salvage (but it depends on what you need to do) is that you may simply stop at the SfM phase, then manually do at least partial manual analisys starting from that data? Let me know if you'd like a temporary key for experimenting a bit!

[feinermo]

Dear Andrea

Excuse my ignorance but what is the SfM stage? what data does it produce?

thanks
ofer

[cam3d]

feinermo - SfM stands for Structure from Motion and is the initial stage which involves camera orientation.

The SfM stage concludes with a sparse point cloud driven by camera positions and their matching feature points - it is followed by MvS (MultiviewStereo) which typically results in a much more dense point cloud. More information can be found here: https://www.3dflow.net/technology/