Hacker News Reader

Comment by jwong_ | Hacker News Reader

Seems like the strobing is intended to defeat the autoexposure compensation. Otherwise the camera might see the brightlight and reduce exposure to compensate. Concentrating the lighting on the face only and staying far away from cameras might help that without requiring the strobe.

View on HN

The leds are placed on the chest -- seems like a miss to not place them also around the hoodie near the face. I would suspect even without strobing the over saturation would work should those leds be surrounding the face.

All of that said, wearing something like this out (unless it becomes super commonplace lol) just screams watch me closely (and easily follow me back to some known origin).

> The leds are placed on the chest -- seems like a miss to not place them also around the hoodie near the face.

Based on the documentation, all that would probably do is allow the effect to work at slightly closer distances.

> The leds are placed on the chest -- seems like a miss to not place them also around the hoodie near the face.

Based on the documentation, all that would probably do is allow the effect to work at slightly closer distances:

> In addition, the LEDs need to be a minimum distance from the camera to be effective, as the cone of light from them needs to be wide enough to overlap with the cameras view of the wearers head. In practice, this is about 12ft (~3m).

Putting them on the chest would help with ergonomics. It would work even if you have the hood off and the LEDs would be more consistently positioned.

FR developer here: that strobing is very smart, as the mechanical mechanism starts to adjust the strobe generates another adjustment. If the strobing is at a random interval, it becomes difficult to impossible to compensate in software. FR does not work without face images.

I would think randomizing the strobing would be effective as well. All of this defeating of auto exposure adjustments makes me think back to the days of Macrovision and how easily it could be defeated as it was not randomized at all. Then again, it was also easily just stripped out of the signal.

With random timing, you're more likely to get one period that's long enough for the auto-exposure to correct it and see the face. Autoexposure takes time because it has to use a feedback loop to search for the correct exposure, so if the brightness changes fast enough, it can't keep up.

you set a maximum on period so that doesn't happen. if you develop something with a known pattern, that can be programed in to be countered. hence my example of Macrovision. randomizing the values means it is much harder to counter.

All the camera would have to do is ignore the brightest parts of the image, then correct the ExposureTime based on that image.

That's what they say, but how much autofocusing is actually happening? In the example video the scene is mostly in focus. My guess is it actually messes with the exposure rather than focus.

Reply on news.ycombinator.com