Most active commenters

    ←back to thread

    509 points nullpxl | 17 comments | | HN request time: 2.415s | source | bottom

    Hi! Recently smart-glasses with cameras like the Meta Ray-bans seem to be getting more popular. As does some people's desire to remove/cover up the recording indicator LED. I wanted to see if there's a way to detect when people are recording with these types of glasses, so a little bit ago I started working this project. I've hit a little bit of a wall though so I'm very much open to ideas!

    I've written a bunch more on the link (+photos are there), but essentially this uses 2 fingerprinting approaches: - retro-reflectivity of the camera sensor by looking at IR reflections. mixed results here. - wireless traffic (primarily BLE, also looking into BTC and wifi)

    For the latter, I'm currently just using an ESP32, and I can consistently detect when the Meta Raybans are 1) pairing, 2) first powered on, 3) (less consistently) when they're taken out of the charging case. When they do detect something, it plays a little jingle next to your ear.

    Ideally I want to be able to detect them when they're in use, and not just at boot. I've come across the nRF52840, which seems like it can follow directed BLE traffic beyond the initial broadcast, but from my understanding it would still need to catch the first CONNECT_REQ event regardless. On the bluetooth classic side of things, all the hardware looks really expensive! Any ideas are appreciated. Thanks!

    1. sspiff ◴[] No.46076613[source]
    I remember seeing some celebrities in the late 00s / early 10s with IR-emitting sunglasses or accessories to flood the camera sensors of paparazzi and make it harder for photographers to get spyshots of them.

    Would this approach work for these camera glasses as well, simply flooding them with so much IR spectrum light that their sensors simply can't see you anymore?

    replies(6): >>46076701 #>>46076704 #>>46076770 #>>46077023 #>>46077069 #>>46078306 #
    2. nullpxl ◴[] No.46076701[source]
    One of my future ideas was to have the detection trigger turning a bunch of IR LEDs on to do just this! I've only tested it a little bit against my phone camera (with around 5 850nm LEDs), but it didn't work super well (fairly bright but not enough to be useful). It did work much better in low-light though. My guess is modern cameras have better IR-cut filters, but like I mentioned I only tested against my phone and not the Ray-bans yet.
    replies(1): >>46076779 #
    3. SamDc73 ◴[] No.46076704[source]
    I heard about similar hats being used during the Hong Kong protests, but most modern cameras filter out IR anyway. Reflective jackets tend to work much better; they basically turn you into an overexposed bright blob on camera.
    4. michaelt ◴[] No.46076770[source]
    Well, there's https://www.nii.ac.jp/userimg/press_details_20121212.pdf

    I think fooling facial recognition systems and CCTV-cameras-at-night is easier than fooling professional photographers. Most photograhers' cameras have IR filters, after all. And nobody's got an LED brighter than the sun.

    replies(2): >>46076921 #>>46076935 #
    5. spacedoutman ◴[] No.46076779[source]
    Have you thought about the potential eye/skin damage you would be causing with IR LEDS.
    replies(1): >>46076981 #
    6. adzm ◴[] No.46076921[source]
    > nobody's got an LED brighter than the sun

    It's low density silly fun but I did see these folk attempt to do such a thing with entertaining results https://youtu.be/m1S1r9I6DN4

    7. Tarmo362 ◴[] No.46076935[source]
    On this topic, is there any benefit of trying to fool facial recognition systems with these type of accessories and or wearables, would the system not just mark you as suspicious and keep an even better track of you

    Of course it is a different thing if these are adopted by the masses

    replies(1): >>46077969 #
    8. card_zero ◴[] No.46076981{3}[source]
    Potentially as much as none, because it's UV that does the damage?
    replies(2): >>46077024 #>>46077827 #
    9. beeflet ◴[] No.46077023[source]
    I have been thinking of a device to thwart license plate readers by dumping a ton of IR and/or visible light on the plate before it gets read.

    Perhaps combined with some reflective coating? Retroreflectors are promising

    replies(2): >>46077815 #>>46078820 #
    10. thih9 ◴[] No.46077024{4}[source]
    I guess IR can be harmful (IR lasers, military grade IR LEDs). But yes, likely not the consumer grade IR LED.
    11. Saloc ◴[] No.46077069[source]
    What about correlating transmitted wireless frames with a LED flashing pattern? If the glasses stream video with a variable bitrate codec over wireless, flashing vs. non-flashing should change bandwidth and therefore frame frequency. However, with searching over all channels this might be quite slow in practice.
    12. delis-thumbs-7e ◴[] No.46077815[source]
    Repo men use those readers to track cars to be repossessed. And as it happens, it is very successful industry these days.
    13. consp ◴[] No.46077827{4}[source]
    At some point it pretty much becomes a microwave. Radiation get absorbed and turned into heat. On a small scale not very helpful or harmful. On a larger scale nice to heat your food with but not your head.
    replies(1): >>46084858 #
    14. wongarsu ◴[] No.46077969{3}[source]
    Usually those systems are set up to track faces and/or people, and ignore everything else. If you get a low-confidence detection of a face that's much more likely to be a dog or a band t-shirt than somebody tricking your system. So you would typically ignore everything below a threshold, not flag it.

    You could train a system to detect these kinds of attacks, but that's a lot more sophistication that these types of systems usually have, and would probably be specific to each "attack" (e.g. those glasses with lights would look completely different than the face paint approach)

    The best defense would be a human watching the raw camera feed, since most of these attacks are very obvious to the human eye. But that's expensive. Maybe you could leverage vision-llms, but those are much more expensive than dedicated face-detection or object classification models. Those typically range from sub-million to maybe a hundred million parameters, while you need billions of parameters for a good vision-llm

    15. RobotToaster ◴[] No.46078306[source]
    That only works against night vision cameras. Most cameras have an IR filter that flips into place when when in daylight mode
    16. fwip ◴[] No.46078820[source]
    Just as a heads up, this is likely illegal in many US states. (Legality is not morality - but it's good to know what the law is before you might break it).
    17. nimih ◴[] No.46084858{5}[source]
    You'd have to be moving pretty fast to red-shift IR radiation into microwaves, I think.