"bullet time" multi-camera photobooth questions

I'm going to post this over in studio and lighting as well as 3d/stereo photography to make sure I gather the right experience, but the DIY forum seemed like the best place to start for these.

Basically I'm planning building a multi-camera photobooth for some upcoming events like this http://gallery.a1array.com/rbmanyc/160514-064451

Have some questions on hardware and software:

1) cameras - considering the gif conversion will downgrade resolution, I'm thinking building this around a 12mp sensor like that in the canon XSi would be plenty. Also, the XSi was well reviewed when it was released about 8 years ago, and can be had for less than a hundred bucks each. Thoughts on this as a camera choice?

2) lens/focal length - I'm thinking 30-35mm would be the max I'd want to go in terms of width to balance between getting a decent amount of people in frame (about 4-6 max) and limiting wide angle distortion. I'll be using studio strobes and shooting at a low enough iso to keep the images nice and sharp so the kit 18-55 set on the wide end would be just about perfect with the 1.6 crop.

3) lighting - trying to decide between continuous vs strobe, but I'll likely stick with strobe because that's what I already own, and will allow a fast enough shutter speed to freeze movement at an acceptable shutter speed and aperture.

4) software - I think this will potentially be the hardest part. There's a company called Breeze Systems that makes software to both control multiple cameras (the XSi is one of the older supported cameras - which is partly why I chose it) as well as photobooth kiosk software that automatically generates animated gifs and creates links. I would rather use software that's pre-made and easily accessible rather than hiring someone to make it for me, but wondering if anyone has other recommendations?

Thanks all!

Sr1racha wrote:

1) cameras - considering the gif conversion will downgrade resolution, I'm thinking building this around a 12mp sensor like that in the canon XSi would be plenty. Also, the XSi was well reviewed when it was released about 8 years ago, and can be had for less than a hundred bucks each. Thoughts on this as a camera choice?

Old camera easily tethered. This isn't a demanding application. I'd use a modern PowerShot, but nothing wrong with your choice.

2) lens/focal length - I'm thinking 30-35mm would be the max I'd want to go in terms of width to balance between getting a decent amount of people in frame (about 4-6 max) and limiting wide angle distortion. I'll be using studio strobes and shooting at a low enough iso to keep the images nice and sharp so the kit 18-55 set on the wide end would be just about perfect with the 1.6 crop.

Ok.

3) lighting - trying to decide between continuous vs strobe, but I'll likely stick with strobe because that's what I already own, and will allow a fast enough shutter speed to freeze movement at an acceptable shutter speed and aperture.

Strobe is a safer bet; keep in mind that lighting a scene for 3D is a little touchy.

4) software - I think this will potentially be the hardest part. There's a company called Breeze Systems that makes software to both control multiple cameras (the XSi is one of the older supported cameras - which is partly why I chose it) as well as photobooth kiosk software that automatically generates animated gifs and creates links. I would rather use software that's pre-made and easily accessible rather than hiring someone to make it for me, but wondering if anyone has other recommendations?

It's pretty easy to hack this sort of thing together, but the Breeze Systems stuff is cheap enough to be a potentially great answer. How would it handle making the animated GIF from multiple cameras?  Would it end-up being the $129 per camera deal? This isn't really hard stuff to write your own script for, but the programming effort is significant if you haven't done this sort of thing before.

Keep in mind there are all sorts of secondary issues. For example, power. The built-in batteries will not last long, so you need to externally power the cameras and control logic.

Sr1racha wrote:

Looks like the T3 is another good option, so I'll prob just mix and match the XSi and T3.

What I'll likely end up doing exposure-wise is doing a slightly longer exposure than needed to give all of the cameras enough time to acquire pre-focus, and then do rear-curtain sync and let the strobe freeze motion (rather than trying to freeze motion with a fast shutter speed). This is also because apparently the multi-camera sync typically works best under 1/100 SS which is not quite fast enough to reduce motion blur

I don't think you have a deep enough understanding of what you're trying to do. You need to make the images as consistent as possible to facilitate alignment of the merged result (careful mechanical alignment helps a lot). Mixing cameras is likely to be a nightmare... zoom lenses aren't a great idea either unless you carefully fix the zooms at the same focal length, and even the tiny movements that happen on cameras with IS (even when it is turned off) can break mechanical alignment at the pixel level. You haven't said what resolution you're targeting, but I'd say anything much above 1MP is going to be touchy on the alignment. You do NOT want autofocus because you need the cameras to agree on the focus point -- in fact, you want them to have enough depth of field to cover the all subjects for the pseudo-3D merge. And of course, you want really good sync -- firing most DSLRs via remote USB control gives quite inconsistent timing; dumb wired-remote shutter firing is often a little better.

Fleets of Canon PowerShots under CHDK have been used for this type of capture many times before, including some very well-known sequences. In my research group, we've been controlling and reprogramming Canon PowerShots using CHDK for a decade or so. At least it is easy to sync CHDK PowerShots relatively precisely (for example, our FourSee multicamera does this). If you think about it, for carefully-controlled favorable exposure circumstances, you'll also realize that the image quality out of $100 new PowerShots is actually much better than you'd get out of the old Canon DSLRs -- and not much different from their current DSLRs (especially if you use lousy kit zooms on the DSLRs). There's even the obvious advantage that you can make the inter-lens spacing closer with PowerShot cameras because they're much smaller than the DSLRs, thus giving smoother depth perception as you sequence between views (computationally deriving intermediate frames to smooth the transitions is possible, but very hard to do well).

The bottom line is that I think it would take my group the better part of a week to hack something together that automatically and unattended (i.e., as a photo booth) did what you want with good quality. Tweaking and making the system look pretty probably would take months.

PS: At modest resolutions, multiple webcams or GoPros would be fairly easy ways to build this too. Apparently, sync of GoPros is also easy. There are also Raspberry Pis with camera modules, etc.

Sr1racha wrote:

The problem with most powershots that I know of (granted I haven't done any research on ones that do or do not) is that they don't have full manual exposure control - and even if there are, they are likely at a price point that wouldn't make sense for this prototype or aren't designed to easily change exposure like with a DSLR.

100% wrong. Even on an $80 PowerShot (e.g., I have a fleet of sub-$100 A4000 and ELPH 115), CHDK gives control well beyond what any DSLR normally provides; you can even do things like set shutter speeds well out of normal range (e.g., many can do mechanical shuttering to 1/30000s!). The only thing close is Magic Lantern on Canon DSLRs (ML started as a fork of CHDK).

There are issues in that you can't get shallow DoF with the PowerShots because their lenses are pretty much at the diffraction limit wide open (hence, there is no mechanical aperture on many), but your application doesn't want shallow DoF.

The kit 18-55 canon IS lens is actually a pretty solid optically. It's not a Zeiss but stopped down to f10-f14 and an iso around 200 will produce great and sharp images.

Ever since I tested the S70 vs. Digital Rebel (and I'm not the only one ), my tests have consistently shown the Canon kit zoom to be slightly outresolved by most PowerShots. You should never use above base ISO on the PowerShots (they are completely ISO-less, or as DPReview calls it, ISO invariant) and there is more manufacturing sample variation in the lenses on the cheaper PowerShots, but the old Canon APS-C DSLRs really are fighting a losing battle in peak IQ against modern sub-$100 PowerShots. The old Canon DSLRs only win in IQ with more awkward lighting and/or better lenses.

You might get a kick out of the overview pages I have posted on two CHDK-supported sub-$100 PowerShots that I've had students use in my Cameras as Computing Systems course: A4000 IS and ELPH 115 IS . For example, here are three versions of the same exposure from an ELPH 115 IS:

Out-of-camera JPEG capture

Completely uncorrected raw from the same capture (CHDK DNG converted to JPEG)

Minimally corrected raw from the same capture (CHDK DNG converted to JPEG)

PS: Take a look at the EXIF info. Note that Canon actually lies about the ISO... the OOC JPEG says ISO 100, but the internals show it's really 93. There are a lot of little details that CHDK makes visible, which might have something to do with why Canon doesn't officially support CHDK nor ML.  

Also the much larger APSC sensor will have better dynamic range.

Peak DR on Canon APS-C and FF DSLRs, before the new sensor tech was introduced with the 80D, is pretty mediocre (about 10.5-12EV). I measure at least 10EV from sub-$100 CHDK PowerShot raws. In fact, according to DxO, before the new sensor tech, the highest DR in a Canon, including all their FF bodies, is the PowerShot G7 X at 12.7EV... which uses a 1" Sony sensor (most Sonys are now 13-14EV). BTW, most PowerShots still use CCDs, not CMOS sensors, which actually helps on DR at the sacrifice of readout speed (no 1080 video capture on the CCDs).

Of course, at anything other than base ISO, the newest larger sensors do win... but that's not relevant for your application. Beyond that, for your application, scene DR should be way less than any of these cameras can handle... and you're talking about mapping things into animated GIFs, where representable DR is a very bad joke indeed!

Also as I've mentioned, it's very easy to sync off camera flash with them through either wired or RF, and the great point brought up previously about slight variances in exposure due to lens and body differences could in theory be controlled with slight differences in exposure.

You mean it is easy to sync a flash with ONE camera. With CHDK, you easily can sync flash and an entire fleet of cameras. In fact, I believe you could even use an external flash (with redeye-reduction preflash) to trigger a fleet of cameras to capture the main flash exposure using CHDK motion detect.

The problem with point and shoots, go-pros etc is that of course they are not designed to sync with off camera flash.

Again, you clearly just don't understand how multiple cameras sync.

I don't have a resolution goal, but this will be viewed primarily on social media which will downgrade and compress images anyways, and primarily on mobile devices so 12mp images combined into a reasonably sized animated gif should be a great start. I've used 12mp images for professional work with an A7S (primarily because it was extremely low ambient light) and though I wouldn't blow them up for print use, 12mp is plenty for fun shots.

An A7S outclasses any of the cameras we've mentioned by a huge margin... in most IQ attributes, so does my original NEX-5. A camera having a 12MP sensor does not imply you'll get a smooth 3D animated GIF by simply sequencing 12MP images from multiple cameras: they need to be aligned. You can get about 1MP alignment quality pretty easily, but much more will take some work. There are various free image stitching tools that might suffice for automatic alignment (try looking at Hugin and related tools); there's also plenty of matlab code floating around (which I'd avoid) and OpenCV.

Will check out other sync software you mentioned. Thanks for the reco - I'd read about it, but Breeze was by far the most mentioned (even though I had my doubts considering how basic the website is).

You have A LOT of homework to do....

BTW, there is something terrible about CHDK: most brand-new PowerShots are not supported for a year or so. It takes that long for the open source developer community, which gets no help from Canon at all, to port to a new model. You have to be very careful to get a camera that is supported by CHDK or you might have a very long wait for support....

Sr1racha wrote:

No offense, but your snarky tone is kinda not very helpful.

Sorry. I just get really annoyed when I put effort into answering a question and then the person I was helping responds in a way that makes it clear they didn't actually read my answer.

... I've read through all of the hacks people are using to make powershots work with CHDK and some of them are solving for issues that problems that don't exist with a DSLR body - from turning every camera body on at once, to triggering, to downloading files (most entry level DSLRs have a dedicated port for triggering, separate from a USB port for downloading files).

DSLRs were never designed for fully automatic operation -- which most compacts are. For example, most DSLRs have no way to control lens zoom from the body. Neither DSLRs nor compacts were designed to run an automated photo booth... but compacts are a much closer match -- especially since that's very much the kind of thing CHDK has been widely used for. You can do this easier with industrial cameras, but cost gets really high quite easily... webcams and GoPros aren't too bad a fit either, but they have more lens and image quality issues.

Basically, what you want to do is photographically speaking easy, with the following hard points:

1. Sync of N cameras and flash
2. Physical/computational alignment of the images
3. Automatically doing everything without an operator

The camera IQ, exposure settings to use, construction of an animated GIF, etc. are utterly simple issues with obvious good solutions. #1 is easy using CHDK, touchy with most other approaches (for example, just because you close the wired remote contacts on several identical DSLRs simultaneously does not mean they fire simultaneously -- or necessarily close enough to catch the same flash... and the USB triggering used by most software is even less predictable). #2 is a pain to get right, and it's the real limit on 3D animated GIF quality (making the animated GIFs themselves is trivial with free software like GIMP or ImageMagick once you have aligned images). #3 is the money pit -- you can spend months or years tweaking the user interface.

The Raspberry Pi camera, using an 8 Megapixel Sony sensor, should be considered, as the system is designed to be programmed by users. The Canon cameras are not.

https://www.raspberrypi.org/blog/new-8-megapixel-camera-board-sale-25/

D Cox wrote:

The Raspberry Pi camera, using an 8 Megapixel Sony sensor, should be considered, as the system is designed to be programmed by users. The Canon cameras are not.

https://www.raspberrypi.org/blog/new-8-megapixel-camera-board-sale-25/

The new 8MP board seems a viable choice in many ways, but it is basically a webcam interface with a very cheap PC-board lens... which is why I listed it as the last option in my post above. Sync of Pis is quite easy, although I'm not sure that would actually sync the cameras (e.g., this on flash sync ). It certainly could be a fun thing to play with....

I was looking at something like this a while back...  my approach was a single camera on a rail ( think a roller coaster for your camera ) using a high speed shooting mode and the PC socket on the camera to trigger the strobe.

The camera is released on an angled section of tube, picks up momentum, then follows a horizontal section of the arc with approximately constant velocity.   A mechanical normally open switch is closed as the camera starts the pass - this triggers continuous shooting.  The first shot is blocked by a fixture that blocks the subject with chromakey green.   This initial image is a marker for the software to know "this is the first frame in the series."

That was the plan.  No need for multiple cameras.   The only limits are on the flash, and the continuous shooting rate of the camera.   Its also simple.   And lends itself to instant processing using an eye-fi card.

OK, so that's the mechanical portion.

On the software side was a linux laptop with an eye-fi target on it.   So the series is shot and is staged on the laptop.

The rest is script magic.

A script uses Image Magick to look for the chromakey frame.   when the script find TWO of these marker frames, it moves the first marker file and the rest of that series to a folder - that's your job.

Once the job is built, a script renumbers the images ( except the marker ) to sequential numbers, e.g 000.jpg, 001.jpg, 002.jpg, ....

Image Magick can be used to convert that to an animaged GIF, or use ffmpeg to create a video clip.

This process basically happens in real time.

I would probably use some cloud service to host these, using the REST API's these offer to obtain a direct URL.   Then dump that into a QR code you display on the laptop.   The "guests" just need to snap a photo of the QR code and they can get to their clip.

For in the field data access, get a wireless tethering 4G phone - this can serve as the hotspot in the field for the EyeFi and your laptop.

Its different from Bullet Time in that it is not multiple simultaneous shots - but its probably good enough to give a similar effect.

Similarly, a camera capable of high speed video ( short shutter "sports" video ) perhaps in 4K would do the same - in that case the speed of the camera travel could be much faster.   Then its more of a video processing effort.

It just a different approach.   There was a guy on Vimeo who did a bullet time video using a suspended 2x4 that rotated the camera around the subject.   It was VERY good.  This is just a more safe version of that.

Maybe I'm late on this, but you should check out FFMpeg http://ffmpeg.org

It's command line software that can do what you're trying to do.