Going Through Motions


You might expect a performance like this to be completely “in the zone”: each throw exactly like the previous one. But it’s not. For example, on the 2nd to last lane on the right, he really hooks the ball into the middle of the pins.

Why would the throws be different? I’d always assumed that the art of throwing strikes was to become a meat robot that repeatedly and relentlessly makes the same perfect move over and over. I also would assume that doing them without a break, like this, would be an advantage because the muscle memory of the successful throw would stay in the short-term memory. I’d expect that to reinforce the tendency for the sequence of throws to be more similar rather than less.

That’s probably why I’m not a world record holder.

I also wonder how they make sure it’s not faked. I’d want to see some sort of consistent something going on in the corner of the frame, that could not be altered without it being obvious. That seems like it might be a difficult problem – what if you had the camera on a steadicam with a lava lamp on an arm positioning it in the corner of the capture. I think it’d be very hard to splice together a sequence without the hiccup in the lava lamp being pretty apparent. Although I suppose you could render-in a complete lava lamp sequence. There must be a minimum number of human witnesses, for such a record to count.

------ divider ------

I notice that he bowls 2-handed; I assume that’s probably a better approach since your arms and torso will form a triangle and self-stabilize. I wonder if the throwing technique is subject to toxic masculinity. There was an episode on one of Malcolm Gladwell’s podcasts about Wilt Chamberlain adopting an under-handed free throw technique for a while, which made him amazingly accurate, but he stopped using it because it was “how girls throw.” At that level of performance, I’d expect people only to care about “how winners throw” but that’s probably why I’m not an avid sports fan.

With regard to the question of fakery: I wonder if there’s a market for “evidence cameras” – cameras with features designed to validate totally that their capture is unaltered. Imagine a camera that did blockchaining checksums inside the video stream – it’d be impossible to alter the video stream once it was captured, but you’d still have the problem of someone pointing the camera at a high enough quality fake and recording that. I did some stuff for Taser back in 2009 or so, regarding cop-cams; that’s a slightly different problem but there’s still the issue of maintaining a chain of evidence. In the Taser case, our solution was to be custodians of the data: get it out of the hands of the cops and into someone else’s cloud control, so it couldn’t be deleted or altered. This could be an interesting technical problem: you could keep and capture a hash-stream of clock ticks, GPS coordinates, compass orientation, and accelerometer data. You’d still be able to fake it if you had the camera mounted so it was recording off a high resolution composite suspended in front of it, and you replicated the movements and timing. I’m pretty sure that having a few human witnesses is better and cheaper. But collusion is a problem, then. (This is all one of the reasons I dislike pro sports: it’s unsporting)

If the scenario I outlined above, of photographing a composite, sounds ridiculous to you: that’s how Blade Runner was filmed. So I’m willing to say it can look pretty good if done by an expert.

Comments

  1. says

    Regarding muscle memory: It’d be useful to see him taking shots at a more normal, considered pace. I’m guessing that the need to rush may have led to him not lining up as carefully as he normally would, and that the different shots are the result of having to make last-moment adjustments in compensation, and that his usual, unhasty style may well be more “meat robot” like.

  2. AndrewD says

    I would assume that, with sufficient funds and equipment, any video could be faked. I also assume that any anti-faking technology only has to make faking difficult and/or detectable. This is really the case for any security- the cost of defeating it should outweigh the gains from defeating it.

  3. says

    Daz: Uffish, yet slightly frabjous@#1:
    I’m guessing that the need to rush may have led to him not lining up as carefully as he normally would

    Aha! Yes, I bet that’s it.

    I sent the video to a friend who’s a bowler and they pointed out that with 5-pin bowling, this sort of strike-throwing pretty much doesn’t happen. I hadn’t thought of the effect of the target pin-density as being particularly important, but of course it is.

  4. says

    AndrewD@#2:
    I would assume that, with sufficient funds and equipment, any video could be faked. I also assume that any anti-faking technology only has to make faking difficult and/or detectable. This is really the case for any security- the cost of defeating it should outweigh the gains from defeating it.

    That’s an excellent summary of security in general: it’s all a cost/benefit analysis, where the defender is trying to raise the attackers’ costs to be unaccepably high. The assumption is always that a failure can be caused by a sufficiently motivated attacker. Where things are interesting is when the cost of attack is extremely low and the defender has assumed otherwise.

  5. hjhornbeck says

    This would be easy to fake. Your first step would be to get a high-quality “camera solve” via PFTrack, which would analyze the video frames and extract the camera’s position and scene depth at each instant. Since the pins themselves are so distant, a photo-realistic render would be easy to generate. Compositing the render back in would be the tricky step, but it helps greatly that the replacement is almost identical to the original. It’s much easier than the Air Force One scene in Iron Man 3, and I dare you to spot a compositing error there.

    I wonder if there’s a market for “evidence cameras” – cameras with features designed to validate totally that their capture is unaltered. Imagine a camera that did blockchaining checksums inside the video stream – it’d be impossible to alter the video stream once it was captured, but you’d still have the problem of someone pointing the camera at a high enough quality fake and recording that.

    Both Nikon and Canon offer image verification via PK. Both systems were cracked in short order. It’s an incredibly hard problem to solve.

  6. says

    hjhornbeck@#5:
    Your first step would be to get a high-quality “camera solve” via PFTrack, which would analyze the video frames and extract the camera’s position and scene depth at each instant. Since the pins themselves are so distant, a photo-realistic render would be easy to generate.

    That made me look up PFTrack and check out their website. Wow, that’s cool stuff! I like the depth-sensing camera and iPad rig. And, yeah, now that you mention it, state-of-the-art graphics and tools probably wouldn’t be necessary. It was laborious, but Stephen Chow had comparable effects in “Shaolin Soccer” back in 2001. (I mention that as an example only because it was a “low cost” production) I guess making high quality sports fakes is probably a non-starter; you’d need to have a semi-credible athlete first. If I produced a video of myself winning the Boston Marathon, you could take one look at me and go “non runner.”

    It’s an incredibly hard problem to solve.

    Yup, and the cryptography is the easy part.

Leave a Reply