Talkin' Truth Beam with Red Batman and the PoliePals

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > YOSPOS > Talkin' Truth Beam with Red Batman and the PoliePals

SubG: Aug 19, 2004; It's a hard world for little things.

MononcQc posted:

I saw the inside of some projects that wanted to leverage ledgers and certificate authorities to deal with surveillance cameras and their main objective was a lot more of being able to prove that from the start of the assembly chain down to deployment in the field, only authorized personnel were touching things. Like making sure nobody in the manufacturing plant adds the wrong parts or counterfeit equipment in a shipment that lets them gain access to a government surveillance network.

other uses came from control-level software for surveillance networks to be able to tie specific cameras and installations to role-based access control and accurate accounting of who has access to which time slices of video/audio files based on mandates, and then digitally signing the content such that you have the ability to trace the whole chain of custody/tempering.

the goal is never to show that other videos are fake or not, but to show that yours is real and was not mismanaged nor tampered. This is okay because for surveillance camera networks, a given camera exists at a location at a point in time and by proving yours is real, you disprove all competing videos for your source and location.

proving that any and all videos are real or fake is loving bonkers and impractical.

and then someone does something equivalent to holding an ipad in front of the camera

the idea of a digital proof-of-life sort of thing isn't a terrible idea on its face, but getting to provable tamper-resistance is a lot more complicated that the proposed "truth beam". which is susceptible to what in effect is a replay attack (if i can paint the thing i want to be verifiable with the flashlight of truthiness, then someone can simultaneously be painting an arbitrary number of alternate scenes with exactly the same signal). i can provide a "solution" to this problem but the bad news is that it involves qm and isn't likely to be ever practical for poo poo like surveillance cameras and cell phone videos outside of a hopelessly speculative science fiction future

# ¿ Dec 15, 2023 00:30

Adbot: ADBOT LOVES YOU

# ¿ May 15, 2024 02:02

SubG: Aug 19, 2004; It's a hard world for little things.

Xathal posted:

Imagine each iteration of the recording being represented by a six-channel matrix - the emission and response. An autoencoder is trained to map this matrix into a latent space. A generator attempts to produce fake examples within this latent space (optionally seeded by noise-augmented real examples), and a discriminator learns to distinguish the reals from the fakes.
The discriminator detects that the ipad does not react to the projector's light emissions like a real scene does. In fact, even if you make super-realistic androids and have them act our the scene, their skin etc. would have to react to light like a human. The analogue hole here is replicants. This is addressed by the more advanced systems (imagine every Truth Beam node being a GPS emitter and receiver, allowing the approximate relative positions of each node to be recorded).

the ipad comment was in response to MononcQc, but sure

do you have a, like, formatted-for-journal-publication paper on this? i don't have a twitter account and the op links a couple twitter posts whose replies i can't (and honestly don't particularly want to) read

Xathal posted:

No. Only the initialisation vector can be projected on multiple scenes (which is fine). Future projections are derived from hashes of the returned camera image. If you simply project the hashes from another loop, they won't correspond to the hashes of the returned images and verification will fail.

okay, cool. this does mean that you're resistant to a simple replay attack, but not a mitm attack. is the remedy for eve intercepting alice's flashlight beam and redirecting it to their own scene supposed to be that alice will notice that her own (the "legit" scene) data fails verification?

# ¿ Dec 15, 2023 23:41

SubG: Aug 19, 2004; It's a hard world for little things.

Shame Boy posted:

ah gently caress i forgot twitter's bad now sorry

twitter's always been bad, op

but no worries. i was just hoping for something more...formal...to review

# ¿ Dec 16, 2023 00:11

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > YOSPOS > Talkin' Truth Beam with Red Batman and the PoliePals