A camera that shoots 40k FPS decided the 100-meter sprint final

I kinda hate this article because it could be so much better and yet it isn’t.

The headline is clearly, in some sense, silly. It’s silly because the usual notion people have of fps is something like video where you need to move your roll of film or read out from your sensor at that rate. The ‘frame rate’ for a photo finish is much more incidental to the way the image is made.

If you have an image from a typical digital camera that is 4000px high formed by a typical two-curtain shutter that shutters in 10ms, you expose 400 rows per ms or, in some silly sense, 400k FPS. If you wanted a 1-pixel high slit to be exposed then that’s an exposure of 1/400000s, which is faster than any camera I’m aware of. But I think the analogy is useful – instead of a shutter moving a slit over the focal plane, exposing different parts of the sensor at slightly different times, imagine the sensor being moved behind a fixed slit and then reading out the values from it. You can reasonably easily imagine doing this with film too – just move it past the slit at a steady rate – not dissimilarly from taking an old-school panorama camera and moving the body instead of the lens.

I think the thing happening here is not a moving sensor but rather a 1 pixel wide sensor (or perhaps a few pixels for colour reasons). This makes it thousands of times smaller than the resolution of the final image so even a fairly typical cmos sensor at 2e9 pixels per second could read 50000 pixel ‘frames’ at ‘40k FPS’. (In practice the number would probably be lower for synchronisation reasons). When your frame is very skinny, that still gives you plenty of resolution.

I don’t like the article because they did some silly arithmetic that produces a big number instead of digging into interesting details, e.g.

- talking about how the system works as a whole (when does it decide to start/stop the image, maybe something about buffering)

- talking about how much light you need and how you get enough

- talking about the optics, how you keep everyone sharp while still getting enough light, how you even focus such a thing

- talking how you make sure the camera is setup fairly (eg perpendicular to the lanes, able to get a good view of all the lanes)

- maybe something about reliability and how you avoid the bad scenario of the system failing when it matters most

Photo finishes are typically done with a line scan camera. It only captures a single column of pixels at a time. So the horizontal axis in the image is actually time, not space. Super cool stuff.

Stray gusts of wind, slight irregularities in the track's surface, other tiny inequalities between lanes or starting positions are probably going to nudge runner performance enough to overwhelm the precision of even a fairly unremarkable finish camera.

If fairness were the primary consideration, then they would use any old junk camera and be generous in calling ties. This kind of theatrical accuracy is there purely so that Omega can advertise.

If a races are decided by such small amounts, like lane 1 being a millimeter ahead than lane 2, then the lanes have to be the same down to a millimeter for it to be a fair race. I wonder if timing will outpace feasible construction before long and we’ll have to say, “lane 1 finished a zillionth of a second ahead, but that makes it a tie because we can’t tell if they started a zillionth of a meter closer.”

As someone who worked closely to the Omega / Swisstiming operations at the Olympics, this is super cool stuff. Congratulations to all involved in being able to deliver when it mattered.

The team also makes really interesting stuff on other sports, such as Beach Volley. Worth checking how it's done.

There was controversy in 2012 at the U.S. Olympic Trials when Allyson Felix & Jeneba Tarmoh finished neck and neck -- basically the cameras at the time couldn't make out the difference [0]. I think that was pretty hard on the athletes, so I'm glad to see we've made some progress.

[0] https://www.theguardian.com/sport/2012/jul/02/jeneba-tarmoh-...

Back in 1987 I started my first job at the Australian DoD. Our group's remit was explosives testing and instrumentation. One of my jobs was to program a hardware controller/sequencer[0] that would trigger numerous equipment: detonators, accelerometer recorders and high-speed analogue cameras. At our facility, we had two bunkers (for in-situ explosives testing), that had a very small window (maybe 15x15cm that was about a foot deep), behind which sat a very big, very expensive, and very fast high-speed camera. That thing could do 10^6 FPS for very short bursts. It had a relatively long lead time to spin up to speed (hence the need for sequencers to precisely kick of the camera and then detonate the subject device at the right moment). I vaguely remember being told that it cost several million dollars (pretty expensive back in 1987). The team that were in charge of that thing also had a couple of Silicon Graphics IRIS workstations, while I was stuck button punching codes on a vertical, rack-mounted keyboard for the sequencer. That camera and the graphics workstations felt like we were living in the future. I've tried to find out more about the camera but honestly all I can recall is how fast it was, how expensive it was, and that the whole rig took about a square meter or two of floor space. But to eventually be able to see an explosion evolving frame by frame was pretty cool. I fully expect given the pace of change, that by the time 40 years has passed, what was once DoD specced "secret" tech, will be consumer level tech - shrunk from m^3 to several mm^3, and cost down by a factor of 10^5. Amazing.

[0] Zilog Z8 BASIC model controller.

Since this camera scans 1D lines, rather than 2D frames, I don't think FPS is the right unit of measure. Perhaps 40,000 lines per second is correct?

would have liked to have seen a better explanation of the red lines in the image. it sounds like those are added by humans to outline the runners' chests; how do they ensure the lines are drawn on the frame exactly at the finish line?

Why is it so difficult for these articles to insert the photo or place the correct link for it?

Here is the image of the camera:

https://www.swatchgroup.com/sites/default/files/inline-image...

Is there a 1-pixel wide display at the finish line that is showing the Omega / Olympic Rings banners?

I used to be a big time operator of FinishLynx equipment stateside, which is the same idea here used by the Omega, it's a finish line camera.

I have often wondered : Why don't they attach a transponder on each athlete's center-of-chest (using some objective definition) and use that to time each athlete? That can be much much more accurate than any camera.

We can still use cameras for visual confirmation but transponders are much more accurate than any camera.

In the top image, where exactly is the "finish" line?

I'm still left wondering how the position of a runner is calculated: Is it any leading part, centroid, leading non-appendage, nose, or something else?

(It's a line scan camera)

It would have to be column per second cause cam sensor is most likely rolling shutter so the top of the frame would be detected later than the bottom but ideally if you have a global shutter cam sensor you could do entire frames per second but I don’t think a global shutter 40k fps camera exists.

The image presented is so distorted it makes everyday interpretation impossible. They'd be better off with a mesh of laser rangefinders and a camera shooting at a meager 1,000 fps. They say the winner was ahead by 5 thousandths anyway.

Hmm -- call me crazy, but I feel like the positioning of the camera might have wrongly decided the race for Lyles. The first torso to cross the line wins.

If you look at the still image in the article, Lyles' right shoulder is leaning forward and visible to the camera, positioned to his right.

But Thompson's left shoulder seem to be leaning forward, but is hidden in the camera by his head and neck. It's possible Thompson's left shoulder is ahead of Lyles' right shoulder, but the image doesn't seem sufficient evidence to conclude on that.

What do other think?

At first I read the title as "40 FPS" and thought "well thats stupid"

Then I realized it's me thats stupid.

> this video is not available in your location

So much for supposed freedom of speech, Elon

There are WAY better slow motion cameras than that.

Ok, what's the underlying storage?

Kishane should have stuck out his tongue. Guess he didn't see Cars?

Maybe my intuition is unusual or wrong, but it seems like a race that's this tight could have been won by anyone. Is being a thousand of a second ahead at a certain point really showing that you are a better or faster runner?

At some point this seems about as fair as a coin toss to me. :)

Edit: I missed that this point had already been made https://news.ycombinator.com/item?id=41230449

Who manufactured this camera ? this artical is puff brand marketing nonsense ? Anyone got any details on camera manufacturer or similar, actual information?

Even elite marathons now are almost always ending in sprint finishes which is a bit dumb but an obvious problem. The women's olympic marathon was won by less than 3 second after more than 26 miles?

Photos no matter how fast still require human objectivity in evaluation.

Instead there should be some digital signal or laser reflection that is precise?

I don't like the rule where it's any part of the athlete's body first to cross the line, it should be the first athlete to ENTIRELY cross the line, no part of them still remains before the line, not even their trailing foot.