Subject in center of orbit is missing - camera symbols in renderings are facing out

[3DjimH]

My scanner orbits the subject with the camera facing inward, running Zephyr Lite V. 4.353 Windows 10.

The final renderings do a great job on the background (actually a mirror image) but the subject in the center of the scanner is missing, and all of the blue camera position symbols are facing outward.

How do I initially tell Zephyr that the subject is in the center of the camera orbit?

Thanks, Jim

[cam3d]

Hi Jim! - I've not seen this before, though I suspect it might have something to do with the way you have acquired your data-set. Can you please share with us some of the original photographs? I assume you were using a turntable - Did you use masks in any way?

[3DjimH]

[3DjimH]

Thanks for your reply. I am using a turntable, camera orbits and elevates from about -20 to +40 degrees thru the session, about 450 images from a Raspberry Pi 5MP camera using raspistill to capture jpgs at 99% quality setting. Also discovered raspistill "sports" setting that reduces background blur by forcing faster shutter speeds.

Since my original post I have improved lighting and image resolution, and reduced the region of interest significantly in the images. At some point in my tweaking iterations, the subject duck started to appear in the Zephyr output and later the blue camera position depictions all turned around to face the inside.

The problem was probably due to my images having much more background information than subject content. Reduced region of interest settings and improved lighting (better contrast) were probably the tweaks that flipped the blue camera symbols to "looking inward", but my non-systematic troubleshooting approach doesn't lead to a crisp correlation of the results. I'm still tweaking and improving the turntable while learning the photogrammetry processes.

I was using Meshroom with some reasonable results, but that was a painful process and this is supposed to be fun. Challenges are good, but the more immediate gratification using Zephyr inspires more rapid learning and development of my hardware monstrosity.

The attachments above are "before" typical ducky image from unsuccessful runs, "after" duck image after tweaking, and the Frankenscanner. The project is typical of my build-before-design process. Stepper motors run the trolley around the scan table and elevate the camera up the... elevator?

[3DjimH]

Which leads to:

What is the "ideal" background for scanning like this? I have plans for panels surrounding the scanner. Should the background be plain, or would high contrast, simple, non-repeating patterns help the photogrammetry process?

Thanks, Jim

[cam3d]

Hi again Jim - Glad to hear you resolved your problem. In regards to an ideal background, when it comes to scanning something feature-poor like your plastic duck, panels of printed perlin noise should work perfectly to re-enforce the camera orientation.

To further improve your 3D geometry reconstruction, consider preparing the surface of the subject with a light dusting of flour or pepper (further increasing features to match between cameras)

As I'm sure you're aware, more light generally a good thing, so shooting outdoors on a bright overcast day will commonly yield better results than indoor shoots without pro lighting gear.
You will reach hardware limits with your 5mpx rpi camera eventually, though it's quite amazing what you can get out of them with a bit of perseverance! - I like your setup, it's looking very cool

[3DjimH]

Some progress...

The perlin noise backgrounds yielded duck models with perlin noise mixed in. Strange!

I've moved to a black background, now to very non-reflective black flocking material. http://scopestuff.com/ss_flok.htm The support for the subject duck is also flocked, and white dots on the support surface seemed to have helped. It's now a different game! Images attached.

I'm getting one half of a duck, pretty good rendition. The output has visible 3D printing artifacts of the duck, so i'm getting reasonably fine resolution. Being the 3D printed duck has the roughness associated with the printing, it is pretty non-reflective. I ran scans with intentional point-source light sources and see no reflections of those in the raw images, and I'm getting progressively better half-ducks as I adjust variables, so I'm not dusting the subject. Yet.

Also, Zephyr is only liking 40-90 of the 470ish images, all views grouped in a region. The camera positions are grouped in a small area rotationally, and only in a limited vertical range. That's shown in one of the attached images.

Lighting is the only variable I can think of that would produce almost exactly half of the subject in the output. I've used several lighting schemes and get varying results of half ducks. I've also used a very flat, wide professional lighting panel mounted on the carriage behind the camera, with ambient room light being the only other illumination. Same thing, always the same side of the duck. I've moved to a different location in the same room, same half of the duck in the output. I'm ready to move outdoors if needed, but after weeks of cloudy skies we now have bright, stark sunshine.

The Frankenscanner doesn't insert position meta data in the images, so Zephyr isn't getting camera position clues other than the magic it uses on the input images. I'm going to rotate the entire scanner 180 degrees to see if that changes the results, a test that indicates that I am running out of tweaks!

I thought that photogrammetry like Zephyr used background information to help construct camera positions, but my best results to date are with the flat black background that orbits the subject opposite the camera. Fair notice, I sell the flocking material for use in the astronomy telescope world. I don't think that biases my results, but I did have it on hand when I started the black background quest at 2 AM.

At lest this isn't boring!

Thanks, Jim

[3DjimH]

[jms]

Hi Jim,

Looks like we are on a similar path here is my setup.



I found using black background means masking is not required.

I've been using a printed base paper with markers on it - I can't remember where I downloaded it from.

I'm using an arduino with a CNC shield and relay board to provide rotation and remote bluetooth shutter control on my smartphone. This is controlled by a Python script.

Is that rig your own design? I like how the Z axis works I was trying to figure out how to do this!

Best Regards
Julia

[jms]

Here's a link to the base paper (scale) I use :-

http://www.stporter.com/wp-content/uploads/2016/03/Photogrammetric_scale_noncoded_markers_plus_2017.pdf

And heres the lady responsible for it :-

http://www.stporter.com/resources/

Best Regards
Julia