Trouble capturing small elements in scene
Hi all,
I am scanning branches for an academic project, and I have a turntable setup (finally). I've gotten a nice Nikon D810 with a Sigma 24mm ART lens. I'm taking 678 photos of the rotating branch from 4 different elevations. The models i come up with now are better than when I was making them with a D7000 with 48 or so photos, but I'm still getting gaps in the models where the branch sections are thin (a few mm in diameter).
My setup has the camera placed about 0.5 meters from the nearest branch section in order to keep everything in focus. I'm trying to avoid the need for manual closeups, as I need to make this a reproducible setup such that field assistants can carry it out. I have the branch set up in front of a green screen, and i process out the green screen prior to RC processing.
Looking at the photos, the thinnest sections are usually 9 pixels wide or so when furthest from the camera, but they're wider when rotated closer to the camera. Having said that, not everything is super sharp due to the distance.
I'm wondering if there might be some settings in RC I could play with that would capture more of these smaller features? Or, any other thoughts for how to improve the results given the constraints?
Thanks much...
Allie
-
Thanks for your thoughts, Steff. The cameras align with no problem - so the base and the targets aren't issues (I process the images with aruco to generate GCPs prior to alignment).
What is your rationale behind increasing the number of "rows", and how do you think that will address the issue of capturing smaller elements in the scene?
-
Thanks Steff. I see what you're saying (though not sure why the relative orientation of object vs rotational axes should matter), and I agree that spreading the photos across more rows might be desirable, but I think that developing some sort of theory about why the smaller features aren't being captured would be important for moving forward first... any thoughts?
-
well because these branches are like long wires, so more rows will help.. adding a different POV for each row..
the space between the cameras on a given row is very small. But the space between 2 cameras, each from a different row is very large..
so more uniform in this case is really going to play an important role in terms of precision.
just to make sure... use high quality in reconstruction -
i could also be that your branches are moving while capturing, so when you are turning the table, the branches will wiggle a bit i think. If the object moves only a bit you will see it in the result. also check how many pixels in this section are visible from camera.
Great setup, minimal but efficient.
-
Thanks all. @Oliver, good thinking about the branch wiggle. I tried to arrange the setup such that there was no wiggle, and had even first planned to have the rotation stop and branch settle before each pic, but the acceleration was making things worse, and the constant speed seemed to keep things very stable. Still, it's possible my eye wasn't picking up the movement.
Regarding feature size, the section that I've highlighted as having gaps is about 5 pixels wide when furthest from the camera, and about 8-9 pixels wide when closest. The image below is from its closest point. Also of note is that it's not very sharp (how can you be when you're only 9 pixels wide). I took these photos with f/16, as I wanted to keep the closest branches in focus, and hence wanted to keep DOF as wide as possible.
I suspect that it's the resolution (i.e. 8 pixels wide) that is giving me trouble. But I'm not really sure. If there is really something about pixel widths, I've wondered about artificially increasing the resolution in photoshop (bicubic resampling or whatever).
I've also included another example of a probelmatic area. To my eye, the image should be enough for RC to grab onto and make some 3D points. However, it doesn't. Maybe this will give more clues about what's going on, and how to remedy it. Thanks again.
allie
another problematic area:
Closeup of the vertical branch section that RC doesn't register:
Please sign in to leave a comment.
Comments
9 comments