Lafodis160: a DIY Large-Format Digital Camera

eit412608 wrote:

If you attached the tripod to the body of the camera and could remove the back of the camera, then you could replace the back of the camera with a screen that you could use to compose. A 4x6 piece of glass for a photo frame that was scratched with very fine grit sandpaper might work for a cheap ground glass composition screen.

That is feasible, although a laser-cut piece of thin plastic is a more likely screen.

To do that with the current version, you'd need to unscrew the rear section of the camera and we'd need to have a tripod mount on the next section. Not fast, but nothing about Lafodis is fast.

This is by no means in its final form, but Lafodis has been progressing. It's a lot more usable now as well as easier to build, so hopefully we'll soon have the full open source software and hardware design release in good form....

The new herringbone gear drive system, using a separate Arduino Pro Micro for the stepper control, is working very well. It's still under $50 parts cost. Typical positioning accuracy over the complete range of motion seems to be within 30 microns, which is way better than the earlier versions did (especially rotationally) and I think is pretty good for 100%-3D-printed motion parts. The drive controls have also been greatly simplified, basically taking feedrate, target absolute angle and radius position, and allowing setting of a new home position.

The ESP32-CAM sensor in use here now also has an NIR blocking filter, so colors are much truer... although the filter is removable for NIR work. A typical single exposure (not even at highest quality setting) is:

A single capture using ESP32-CAM with Wollensak 135mm f/4.5 @ f/11

Really quite reasonable for 2.2 micron pixels in a $2 OV2640 camera!

Aiming the camera and tweaking image parameters is now done via an HTML interface accessed using any browser via 802.11 WiFi. Commands and image capture download is also now via WiFi rather than BlueTooth, although the TF card slot is also available for local capture storage (up to about 20K 1600x1200 ESP32-CAM captures per 4GB card).

It was a VERY windy day today. In fact, just after a 16x16mm capture (~52MP), the wind blew the tripod Lafodis was sitting on over, smashing the lens into the ground hard enough to shatter the lens mount plate -- I'm printing a replacement with an improved design. Nothing else was damaged. Anyway, the point is that the wind definitely was moving the tree branches around a lot, which makes it a really good test of dynamic capture/stitching.

Here's the crude first light stitch from that 16x16mm capture. Remember that Lafodis full capture area is a 160mm diameter circle, so this is just 0.3% of a full-resolution capture for Lafodis....

First light stitch, done in real time on an old laptop during Lafodis capture

The scan order used for this was a modified Hilbert walk, which improves the temporal adjacency of spatially nearby samples.

The blending in this test is rather crude and the scan parameters fell slightly short of covering the full space (which is why the black lower left corner), but it worked pretty well. Just for comparison, Hugin failed to automatically stitch these 196 image captures after using more than 40 CPU hours, so the new confidence-based stitch algorithm for Lafodis is working way better to produce this incrementally in real time during capture. Fully automatic stitch quality should be improving significantly within the next month or so as the blending, etc., get tweaked.

I'll be posting more updates as things develop. It's been really slow getting this tweaked to be more user-friendly for open source release during the pandemic... but everything is going slowly during the pandemic. 

As you might have guessed from the not-horrific 1st light image I posted, things are getting much cleaner now with Lafodis. Here are the two latest versions:

The newest is the one on the left. Literally 100% of the mechanical parts are 3D printed, including the screws. Here are the latest tweaks:

• There is a 3D-printed internal filter holder for an NIR blocking filter.
• The drive system has been tweaked for higher accuracy. The Herringbone gear drives on the one to the right were a big improvement over the original radius worm gear and direct angle drive, but the latest version adds things like a full track for the internal ESP32-CAM mount rather than counting on the Herringbone gear pressure alone to prevent wobble. In addition, nearly all fitted parts are now modeled more precisely to improve fit (at the expense of taking longer for OpenSCAD to make STL files from the 3D models).
• There are now wire routing channels inside to keep the internal wiring a bit more tidy.
• The printed clamp that holds the Arduino Pro Micro in place now has an integrated light shield to keep the standard lights on the Arduino from creating stray light inside the camera.
• There are now parametric options to support lenses with very short or very long rear focus. The (interchangeable) focusing mechanism was originally tuned for maximum focus range with lenses around 135mm -- typical "short normals" for 4"x5" film cameras. The Lafodis on the left is shown tuned for around 90mm.
• Rather than the lens mounting directly in the 3D-printed front plate (which would still be an option), I've now switched to preferring a lens mount that accepts standard 4"x4" lensboards... which the blue and pink things actually are (see my old Instructable Custom Lensboards for a Large-Format Camera ). Just for comparison, here's that blue lensboard mounted on my old B&J:

• There is now a computed simulated leather texturing (ah, math ) on two key parts of the camera body to improve grip and help hide the usual 3D-printing line artifacts. The 2nd version (the first Lafodis160) relied on hammered bronze spray paint to give a nicer exterior finish, and that was fine, but the fake leather texturing doesn't need painting to look nice.

Although printing time and weight will vary significantly depending on slicing, material, and printer used, it takes about 24 hours to print a full set of parts on my $180 AnyCubic Linear Plus. No single part took more than 6 hours. The materials used were black PLA, and silver and copper silk PLA. The full set of PLA printed parts weighs less than 600g, which means the material cost of the printed parts is around $12. Of course, somebody else making it for you will charge significantly more than that because it is using a printer for a full day and requires some attention (i.e., you can't fit all parts on the platform simultaneously to do it all in one printing run).

Anyway, I'm really hoping to have a full public domain release by January....

Of course, I am open to comments & suggestions. I already have plans for later versions to use:

• A larger sensor.
• A tilt/shift lens mount.
• A second ESP32-CAM as a remote viewfinder.

However, first things first... I'm still trying to tweak everything to be easy to make and use....

Graham Gibson wrote:

Cool to see your progress on this project.

It's way behind schedule, so I wanted to let folks know it's still coming....

What are the printer bed size requirements?

Largest part is the outer shell which includes the rotational Herringbone drive track and a bunch of 1/4-20 threaded tripod mounts. It prints as one piece within a 188mm diameter by 100mm tall build area.

That's a tight fit on consumer printers, but it fits most. It is just a hair too big for the 180x180mm of a Prusa MINI+, but most 3D printers have beds that can handle at least a 200x200mm square. For example, my MakerGear M2 build volume is 200x250mm by 200mm tall. It should also fit on Kossels; e.g., my AnyCubic Linear Plus is a Kossel with a 230mm diameter by 300mm tall build area.

BTW, I'm pretty proud of the fact I'm getting a tad more than a 160mm diameter image circle captured in a housing that contains all the electronics and mechanics and is only 188mm external diameter. There's no way to get close to that with a more conventional rectangular housing.

Do you have some recent images to share since last week?

Not yet. I've been too busy to test it yet. It's the end of the semester panic, plus I'm presenting unrelated work at a conference... but that didn't stop me from printing the new parts.

How's it going with the stitching? There's still a pretty obvious mosaic going on in the last image shared. I assume this is due to the sensor micro-array as I mentioned earlier in the thread. Have you figured out how to overcome that?

The stitching quality issue is a combination of poor blending and that I'm doing it completely incremental right now, which means that positioning errors can accumulate. The right way to do it is basically to do global alignment constraint solving for the final stitch, but still use the incremental version to confirm coverage with minimal motion artifacting. I'm hoping the tighter positioning accuracy on the latest version will help the incremental stitch quality.

The color/shading shift off axis is not as severe here as you found, and that's partly because I don't use the whole sensor image area for the stitch -- I use a 1200 pixel circle centered in the 1600x1200 pixel sensor. For what it's worth, I also do confidence weighting with confidence decaying as you go further off axis within the sensor. I also found that the off-axis behavior was much worse if I wasn't careful about how the NIR blocking filter was installed; it's now in a little 3D-printed frame that keeps it precisely parallel to the sensor and prevents edge reflections from the filter glass.

Next priority is actually speeding-up the capture. In theory, I might be able to get close to 5MP/s, but it's still way under 1MP/s due mostly to WiFi image transfer.

spiffariffic wrote:

Following

It's still coming....  These things often take much longer than they should, and the pandemic hasn't been helping....

ProfHankD wrote:

spiffariffic wrote:

Following

It's still coming.... These things often take much longer than they should, and the pandemic hasn't been helping....

Subscribed!

I just had to "part out" a large Ricoh office printer/copier, and have a box full of stepper motors!

fferreres wrote:

I think I had asked, but maybe I imagined. Would this work adapted as a pure Graflok 4x5 back? with no front at all in it?

Not really, because it would need to stick out a lot -- the back would be large, actually about the same size as the entire Lafodis. There's no realy advantage to doing that. I can replicate the movements of a flexible 4x5 more easily and compactly as a new unit than by trying to graft a scanning back onto an old camera.

There is still some potential for a 4x5 or larger sensor to effectively use old 4x5 cameras, but even then, a fully new body can have lots of additional capabilities.