26 May 2023

For my cousin’s wedding I did not make a photo booth but a video booth. – With an array of DSLRs to create a bullet time effect.

Since this is a very visual project, you should really watch the video. This blog post contains the same info with more details, but those details are probably only relevant if you really want to recreate this project, which is not an easy thing to do.

What is a bullet time video booth?

The story of this project actually begins in 2017 when I created a video booth for my own wedding. At many wedding receptions you can find photo booths, which are simple camera setups with a remote trigger to allow guests to create some memories of the special day. Usually this involves silly props like hats, wigs and giant glasses and it is as much about creating photos as a memory as it is about the fun of being creative while doing so. …and for my wedding I mixed it up a bit by putting my Sony NEX-5T on a tripod and allowing guests to take short 5 second clips¹ instead of static photos. Later I cut all the clips into one long video with upbeat music and we still enjoy watching this memory today.

My family enjoyed it so much, that one of my cousins used the same setup a few years later at her wedding and when another cousin planned his wedding for this year, he also asked for a video booth. But this time I wanted to try something new by adding a bullet time effect.

In case you never heard of it: The bullet time effect was famously introduced with the movie “The Matrix” from 1999. It really became a signature style for these movies and has been copied and reused so many times that it pretty much became a movie cliché. But while it can easily be recreated in purely rendered form like animated movies or video games, it is still quite involved and costly to achieve with real cameras: You set up an array of photo cameras, trigger them all simultaneously² and play back the individual photos as frames of a video. The result is that the scene seems to be frozen in time while the camera moves around freely.

And here you see the problem for hobbyists. If you are not rendering the entire scene but need to take photos of real people, this can quickly become expensive. Bullet time-time costs one camera per frame. For someone shooting with the typical “European” 25 fps³ that’s 25 cameras per second. And that of course also explains why my bullet time video booth only uses this effect for the transitions instead of recording several seconds of bullet time footage. I simply wanted to keep the costs for this project in a reasonable range by only using twelve cameras.

This means that I only have a bit less than half a second of bullet time footage per guest. But as you could see in the little example clip above (and more in the Youtube video) the effect is still nice and appears longer. To achieve this I used two tricks:

• I trigger the cameras at the beginning and the end of each video recording and they are arranged in an arch towards the wall behind the guests. By mirroring every second guest clip and playing the bullet time photos in different order I can combine the two twelve photo sequences from the end of the previous guest clip and the beginning of the next guest clip to make it look like the camera spins through the wall to the next clip. An additional blurred frame at the location where the camera would be inside the wall and a non-equidistant placement of the cameras for an accelerating motion towards the wall help to sell the effect. This adds up to 25 frames and therefore one second of bullet time footage.
• I created the final video of all guests (and the example above) in DaVinci Resolve and used its Optical Flow Estimation to generate artifical additional frames. This allows to smoothly slow down the bullet time effect further and also gradually slow down the video recording after leaving a bullet time transition and before entering the next bullet time transition.

So, I was able to pull this of with twelve cameras plus the one for the video recording. But this setup requires more than just the cameras…

Setup

In this section I will go through the important components of the bullet time video booth. Keep in mind that pretty much everything was selected to keep the costs as low as possible. Unfortunately, this also means that some components were a bad decision in hind sight and some components are simply more expensive or better equipment which I happened to already have. So, if you want to build something similar, be prepared to pick your own components and adapt the software to your needs.

Bullet time camera array

Let’s start by looking at the part that obviously is the most expensive because that purchase comes with a factor of twelve: The cameras used for the bullet time effect.

When I said that the bullet time effect is still expensive, I have to admit that it has become much cheaper since 1999. Cameras have become so readily available and good today that a comparison to 1999 equipment would be ridiculous. Still, I set myself a limit of 50€ per camera, which means that the twelve cameras alone would cost 600€, which I find quite heavy for a fun gimmick to a video booth⁴. And now think about what cameras you can get for 50€… There are some possibilities like webcams, used smartphones, used action cams, etc. But used ones for 50€ will have poor image quality and their small sensors are not ideal for this effect as they require more light for short shutter speeds or you will get a lot of motion blur. Raspberry Pi cams are also an interesting alternative in the maker world, but the cheap ones have very poor image quality and the better ones cost 50€ alone without a lens – and you need a few actual Raspberry Pis which are even more expensive.

Instead, I looked for old used DSLRs. They have large (albeit old⁵) sensors, they support means to remotely trigger the shutter, they have plenty of resolution, their settings can be controlled precisely and nobody except me is interested in them anymore. The trick is that we are using the photo mode of these old DSLRs to create a video, which means that we are comparing old photo specs with modern video specs. So, I looked for the oldest mainstream beginner DSLR that has a decent resolution and picked the Canon EOS 400D⁶ from 2006.

That camera is now almost 17 years old, which means that craigslist⁷ is full of those cameras. There are so many people trying to sell their old camera with almost nobody being interested in such outdated technology⁸ that I could simply work through the offers starting with the oldest listings. I offered my 50 bucks for the 400D and its mediocre kit lens, reminded the seller that nobody has even looked at the listing in six months, and either he accepted or I contacted the next one in the list.

But is it good enough? Well, that’s where we can now compare the photo specs to the video specs. The old 400D from 2006 has a photo resolution (it cannot even shoot video) of 3888×2592 pixels, which is more than our modern 4k (or actually UltraHD) video with 3840×2160.

Video camera

But since the Canons cannot record video⁹ I need one more camera as the main camera that shoots the video. Initially, when I selected the 400Ds, I aimed for 4k video and planned to use my trusted Sony a6400 for this. Unfortunately, it turned out to be rather difficult to trigger the video recording precisely and transfer the resulting video file. Both can be done via USB, but one requires the camera to be in control mode and the other one requires it to be in mass storage mode. Video files apparently cannot be transferred in PTP mode, which seems to be an oversight by Sony.

I either needed to use a Wifi SD card like I did in my old original video booth or record externally with an HDMI grabber. Since transferring 4k videos with a Wifi card (or maybe even recording to that old Wifi SD card) seemed like a bad idea, I dropped the 4k goal and went with one of my cheap 1080p HDMI grabbers instead. And since I did not need to tie up my good camera in this project if I only record 1080p footage through a mediocre HDMI grabber, I used my Sony a5000 instead. Some of you might remember it from an older post, because it requires a hack to get a clean HDMI signal.

So, we are down to FullHD, but as a plus using an old cheap camera gives me some valuable peace of mind when leaving the bullet time rig unattended at the wedding venue.

Quarter circle stand

At that point I thought that I had bought the most expensive part of my bullet time rig – until I faced the question of how to mount the cameras. Whichever solution comes to your mind right now: Remember to multiply its cost by twelve and think again.

Using twelve individual super-cheap tripods is impractical as someone will easily hit one of the tripod feet and misalign a camera from the other ones. So, I was sure that I need a common stand to which I mount all cameras. I first thought of traverse material as found on stage, but tha