A blank wall can show how many people are in a room and what they’re doing | Hacker News Reader

A blank wall can show how many people are in a room and what they’re doing

scientificamerican.com

81 pointsisaacfrond14 comments

I recall that when the first Apple stores were opened that they used an overhead camera technology that recorded customers moving about the store. This included customers that walked up to the window, and those that walked inside. The images were rendered as black silhouettes so as to anonymize the customers, but the technology helped Apple understand traffic patterns in their stores.

It's unclear to me why the wall would need to be "blank". Maybe simply unobstructed and flat or moderately consistent reflectance Lambertian? A dark black wall would be noticeably more difficult, while one painted with several bright colors could be corrected fairly easily. Even a wall with fixed fine black on white patterns (e.g. text) should be comparable, while a white sheet with fans blowing on it incredibly difficult.

It’s really a statement about what the researchers actually did for this paper, not a statement about what’s required or possible in general now or in the future, right?

Because they qualify this is for blank walls both in the abstract and in the video, I’d first assume it’s possible they tried it and ran into substantial issues - even though it might logically seem easy. Maybe it will take a bit more math or engineering to handle non-blank walls, and it’s being saved for the next paper. Since non-blank walls will increase the dimensionality (and training time) of the problem, and likely come with reduced effectiveness in some cases, and since a result with blank walls is good enough for publication, maybe this is just a first step?

>It's unclear to me why the wall would need to be "blank"

The initial state of the wall (before people are introduced) should be known.

With a solid color wall you can find shadows of people, objects, etc.

But with a wall with objects on it, you need an image of it before the room was filled, in order to interpret the shadows. (From a machine standpoint. As humans we can process out the art or whatever is on the wall)

Pretty sure Plato wrote about this in his _Allegory of the cave_ (c 400BC),

And the actual work including papers and video is here: http://wallcamera.csail.mit.edu/

I still don't understand how they're able to get a detectable optical signal through brick and concrete walls. It'd be nice if they actually measured the transmissibility of said walls in the paper.

It's not through the wall, it's from reflections off the wall. Imagine being able to see around a corner because all walls were mirrors, except this works for walls that are not traditional mirrors.

My understanding is that these are all super fragile though. All these experiments are done in super controlled environments. Sounds cool and might be a step forward but never going to be applied realistically for a while I think

See also Inference from inadequate and inaccurate data [1]

[1] https://pubmed.ncbi.nlm.nih.gov/16591812/

This reminds me of trying to explain Ambient Occlusion https://en.m.wikipedia.org/wiki/Ambient_occlusion

Someone using their phone will throw this off too I'm guessing...

Is it any more advanced than this?

https://youtu.be/H5xmo7iJ7KA

This work used regular cameras to see through brick and concrete block walls.

Reply on news.ycombinator.com