Lafodis160: a DIY Large-Format Digital Camera

I'm a Professor of Electrical and Computer Engineering at the University of Kentucky and one of the three research papers I will be (virtually) presenting at Electronic Imaging 2021 is An Ultra-Low-Cost Large-Format Wireless IoT Camera. You'll have to wait for the conference for all the details and the open-source design.... However, I've reached a significant milestone in development of the second prototype: collection of "first light" images from the assembled system. So, here's a taste of what it is that I'm working on....

The series of cameras are called Lafodis: LArge FOrmat DIgital Scanning. The first was Lafodis 4x5, and you can probably guess what 4x5 refers to. This second one is Lafodis160. The "160" part refers to the idea that it is designed for lenses with image circles up to 160mm diameter -- basically somewhat oversize for 4"x5" format. Some current quick specs (which of course may change without notice):

• Resolution: targeting 500MP @ 4"x5"; theoretical peak resolution is 2.6GP
• Dynamic range: 8-10EV; theoretical HDR limit is 20EV
• Color: RGB CFA, no integrated NIR filter
• Scan speed: currently <1MP/s; theoretical peak ~10MP/s
• Construction: 3D-printed body, linear rail, drive screw, electronics mounts, lens extension, lens focus thread, and lens mount plate
• Dimensions: approx. 171mm diameter, ~190mm deep with 135mm lens
• Weight: 877g including Wollensak 135mm f/4.5 enlarging raptar
• Electronics: ESP32-CAM and two 28BYJ-48 steppers with ULN2003 drivers
• Capture control: wireless via BlueTooth, C++ program using OpenCV
• Firmware update: wireless via wifi, Arduino OTA compatible
• Power: 5V via USB connector from external source
• Build materials cost: approximately $50 without lens
• Build equipment/skills needed: 3D printer with at least 180mm diameter by 120mm tall build volume, some wire-wrap & soldering required

What's it look like? This:

It's not small, but it's surprisingly lightweight and, much to my surprise, it doesn't seem to have major vibration/resonance problems even when used on a very flimsy tripod.

I'm not going to show you the assembled first light images yet; the stitching didn't work as it should have and there's some really annoying dirt on the sensor that makes it look even worse. However, here's a monochrome 1:1 sample from a single capture at the 500MP 4x5 resolution:

1:1 monochrome sample at 500MP 4x5 resolution

It's not awesome, but the detail is pretty good for off-axis on a cheap old 135mm enlarger lens stopped down to f/22. Of course, f/22 is well past the diffraction limit for this resolution, but depth of field is so thin that I felt it necessary to stop down for this test shot, which was literally some trees photographed through the window of my study.

Anyway, that's enough of a teaser for now.

This has been a significantly harder device to develop than was expected, but build is a lot easier for Lafodis160 than it was for Lafodis 4x5. It's pretty much good enough now as a research prototype, but there are definitely some additional simplifications and improvements that would make sense. So, I could follow this with development of a third version, and it might even be the start of a whole series of open source DIY large-format cameras... or not -- which is most of why I'm posting this now.

I'd like to get a feel for how many people would want to make and/or use such a wacky camera? The question I'm trying to answer is really how much more effort I should put into making this into an open source design more accessible to others?

ProfHankD wrote:

I'd like to get a feel for how many people would want to make and/or use such a wacky camera? The question I'm trying to answer is really how much more effort I should put into making this into an open source design more accessible to others?

I would, thanks.

Sterling
--
Lens Grit

ProfHankD wrote:

What's it look like? This:

This looks awesome and I like the color scheme.  The whole setup reminds me of Celestron C-90s.

I'm not going to show you the assembled first light images yet; the stitching didn't work as it should have and there's some really annoying dirt on the sensor that makes it look even worse. However, here's a monochrome 1:1 sample from a single capture at the 500MP 4x5 resolution:

1:1 monochrome sample at 500MP 4x5 resolution

It's not awesome, but the detail is pretty good for off-axis on a cheap old 135mm enlarger lens stopped down to f/22. Of course, f/22 is well past the diffraction limit for this resolution, but depth of field is so thin that I felt it necessary to stop down for this test shot, which was literally some trees photographed through the window of my study.

Not bad at all considering there'd be 1350-ish more crops of this size in a 500MP image!  Does it hold up at the pixel-peeping-level? Perhaps not, but after downsizing and some tasteful sharpening, there wouldn't be much to complain about.  At something like 60MP things would look great!

I'd like to get a feel for how many people would want to make and/or use such a wacky camera? The question I'm trying to answer is really how much more effort I should put into making this into an open source design more accessible to others?

It would be a very cost-effective way to satiate my curiosity about some of the medium and large format lenses I've come across.

Tons o Glass 0 Class wrote:

ProfHankD wrote:

What's it look like? This:

This looks awesome and I like the color scheme. The whole setup reminds me of Celestron C-90s.

The color scheme is basically anything you want -- provided it's light tight. Everything here has at least one coat of Black 2.0 on the inside and/or outside to ensure that. The lens mount is printed in copper silk PLA. The rest is black PLA, but the main body outside was spray painted with hammered bronze over the Black 2.0 to give a nice & very durable finish. This is a prototype for a research paper, so I wasn't about to make it in Fisher-Price colors, but one could....

I'm not going to show you the assembled first light images yet; the stitching didn't work as it should have and there's some really annoying dirt on the sensor that makes it look even worse. However, here's a monochrome 1:1 sample from a single capture at the 500MP 4x5 resolution:

1:1 monochrome sample at 500MP 4x5 resolution

It's not awesome, but the detail is pretty good for off-axis on a cheap old 135mm enlarger lens stopped down to f/22. Of course, f/22 is well past the diffraction limit for this resolution, but depth of field is so thin that I felt it necessary to stop down for this test shot, which was literally some trees photographed through the window of my study.

Not bad at all considering there'd be 1350-ish more crops of this size in a 500MP image! Does it hold up at the pixel-peeping-level?

Yes; what you're seeing is the approximate pixel-level contribution of one of thousands of images being stitched.

To be technical, they don't get stitched in the traditional sense. They actually get merged based on computed pixel value certainties... which sounds complex, and sort-of is, but it's really a very similar algorithm to what we published at EI2020 in Senscape: modeling and presentation of uncertainty in fused sensor data live image streams. There's a WWW page on that: http://aggregate.org/DIT/PAINT/

Kinda looks like a "Quaker Oats" box camera!

A great project! Any chance that the sensor can record infrared?

Would love to see more details on both cameras!

KudzuLite wrote:

A great project! Any chance that the sensor can record infrared?

Near IR, yes. True IR, no.

Would love to see more details on both cameras!

You will. The EI conference is in January....

I don't know if this is a project that I would take on but I do think this project being available will make the world a better place.  Please accept my fervent encouragement for your work.

Terrence Miele wrote:

I honestly don't have a clue what you're doing but I find it absolutely fascinating and definitely look forward to more of how this works. I was reading another Mamiya thread trying to figure out what you guys are talking about using scanners to make images attached to a camera or something I don't know sounds interesting but above my current thinking level but I look forward to these experiments

Not long to wait now... Electronic Imaging 2021 has started, and my presentation on Lafodis is at 13:00 EST Jan 19, ISS-070, An ultra-low-cost large-format wireless IoT camera. Thus, you can expect I'll have details posted within a week or two after that.

BTW, I have seen enough interest here and elsewhere that I am working on an improved version (3rd complete system prototype). The main change is that the new version will use a herringbone gear drive for angular movement rather than direct coupling to the stepper drive shaft. That should make motion more precise and also will allow for all wiring to be on the rotating internal platform, whereas the current version has a few wires that could become tangled if an incorrect sequence of motion commands were issued to the camera.

nt = no text

mapachebasura wrote:

A bit late but this looks REALLY interesting. I very much missing having access to medium format equipment and while this looks like a somewhat different beast, may be the interesting thing I'd be very into. Was there any movement on this?

Yes.

The Electronic Imaging 2021 presentation was January 19, and is posted as PDF slides at http://aggregate.org/DIT/iss070.pdf .  Here's what the poster itself looks like:

The final version of the complete paper will be off for publication on Monday, Feb. 8.

A website for everything will also be up very shortly after that. Once that's up, I'll be posting more about it here too.

However, the host software is still in pretty rough form and I am also working on a revised version of Lafodis160 that will use a herringbone geared angular drive rather than direct drive, with a variety of other minor modifications to simplify the system for others to make. At this stage, I'm pretty convinced it is worth making Lafodis something that can be widely used (rather than just a prototype to further research goals), so I'm putting effort into that.

mapachebasura wrote:

Awesome! What do you see as the ultimate end goal of this? Is it purely for it's own sake? Is there an overall objective/ideal use?

Lafodis160 was really intended to be a testbed for some very strange temporally dynamic scan order stuff that really relates to a custom sensor I've been hoping to get built for a LONG TIME now.

However, I now think this is really going to be useful for quite a few photographic tasks, so I want to get it out there. I'm a big believer in digital large format, and this might be dirt slow, but it's also dirt cheap, easily environmentally sealed, and designed for remote operation -- that's not a space any other large-format cameras are really in. It's just a lot of work to take a research prototype to the level of being a reasonable DIY project that would be usable for non-researcher photographers....

mapachebasura wrote:

Is this scalable or easily adjusted to better sensors?

Better sensors tend to be problematic because the cheap ones tend to integrate lens AF, so removing the lens breaks 'em. However, nothing wrong with having multiple OV2640 to speed-up capture. 3 would fit easily, although that would make the scan logic even more odd.

Like, if you wanted to just do a similar project for medium format but with a more expensive (but still relatively cheap!) digi cam. Or would that require entirely new build and software?

I don't see much advantage to much a smaller image circle because the motors and circuitry don't get smaller and image quality competition from FF is closer. Larger can make a lot of sense, but breaks 3D printing with cheap printers.