A Blank Wall Can Show How Many People Are in a Room and What They're Doing (scientificamerican.com) 23
Stare at a blank wall in any room, and you are unlikely to learn much more than the paint color. But a new technology can inconspicuously scan the same surface for shadows and reflections imperceptible to the human eye, then analyze them to determine details, including how many people are in the room -- and what they are doing. From a report: This could be used to spy on activity from around a corner, learn more from a partial view of a space or watch someone avoiding a camera's line of sight. As people move around a room, their bodies block a portion of any available light to create subtle and indistinct "soft shadows" on walls. Brightly colored clothing can cast a dim, reflected glow. But these faint signals are usually drowned out by ambient light from a main source. "If we could do something like subtracting this ambient term from whatever we are observing, then you would just be left with camera noise -- and signal," says Prafull Sharma, a graduate student at the Massachusetts Institute of Technology. Sharma and other M.I.T. researchers isolated that ambient term by filming a wall in a room as its occupants moved around and averaging the frames over time.
This eliminated the shifting shadows cast by the humans, leaving only the light from the main source, plus shadows from furniture or other stationary objects. Then the researchers removed this term from the video in real time, revealing moving shadows on the wall. Next, Sharma's team recorded blank walls in several more rooms in which the researchers enacted various scenarios and activities. Groups of one or two people moved around outside the camera's view. Others crouched, jumped or waved their arms. Then the team fed the videos into a machine-learning model to teach it which soft shadow patterns indicated which behavior. The resulting system can automatically analyze footage of a blank wall in any room in real time, determining the number of people and their actions. The work was accepted as an oral presentation at the 2021 International Conference on Computer Vision in October. Although this system can function without calibration in any room, it performs poorly in dim lighting or in the presence of a flickering light source such as a television.
This eliminated the shifting shadows cast by the humans, leaving only the light from the main source, plus shadows from furniture or other stationary objects. Then the researchers removed this term from the video in real time, revealing moving shadows on the wall. Next, Sharma's team recorded blank walls in several more rooms in which the researchers enacted various scenarios and activities. Groups of one or two people moved around outside the camera's view. Others crouched, jumped or waved their arms. Then the team fed the videos into a machine-learning model to teach it which soft shadow patterns indicated which behavior. The resulting system can automatically analyze footage of a blank wall in any room in real time, determining the number of people and their actions. The work was accepted as an oral presentation at the 2021 International Conference on Computer Vision in October. Although this system can function without calibration in any room, it performs poorly in dim lighting or in the presence of a flickering light source such as a television.
Dont tell Amazon (Score:2)
If you tell Amazon they'll put this in all of their 'smart' devices.
Re: (Score:2)
Re: (Score:2)
You might not be following the cues, so Alexa needs to know what you are actually doing so that she can properly influence your next move. See for Alexa to do her best she needs a continuous stream of everything in your life, seen and unseen, heard and unheard.
Normally if someone wanted this level of information about your life you'd call them a creepy stalker and a criminal. But when your stalker is a cute talking speaker you call it 'smart'.
Re: (Score:2)
How much business has Amazon done purely on the "people who bought this also bought"? I would think a lot, because it's still there, even though it doesn't seem to work that well most of the time.
Re: (Score:1)
People are surprisingly suggestible and don't need many or strong cues.
Especially capitalists. Dangle a dollar bill on a fishing line in front of one and it will be a trip to the ICU to remove the hook, paid for who? Guess who. You.
Re: (Score:2)
"Dangle a dollar bill on a fishing line in front of one and it will be a trip to the ICU"
or a lawyer's office...
https://www.npr.org/2021/08/25... [npr.org]
The researchers also... (Score:2)
Privacy (Score:2)
That's why I cover my walls with hexagonal RGB lights.
Plato's Cave (Score:2)
Re: (Score:1)
Re: (Score:2)
The big difference is that this research assumes the camera is approximately in the same room as the people.
Imagine that: If a camera can see the same space that people are in -- and those people are exaggerated enough in their movements -- the camera can figure out what the people are doing. I am shocked, shocked! Next they may be warning us of the dangers of putting mirrors on walls and/or ceilings.
Re: (Score:1)
So this means we can point a camera at the moon to deduce what people on earth are doing!?
Time to plant more trees!
csw
Re: (Score:2)
Two sides of the same coin (Score:1)
I remember being told about 5 years ago about a similar technical approach: the single pixel camera.
The idea is that the time variation in the single pixel can tell you about patterns of activity without giving away any identifying information. The idea would be you could put it in a hospital or nursing home bathroom or bedroom and be able to tell if the patient has fallen or lost consciousness without him wearing a monitor and without having to put a camera in a private location.
This is the same idea...but
Fuck Scientific American. What happened to you??? (Score:2)
Re: (Score:2)
Looked a lot cleaner than a lot of sites to me.
Also, I am going to guess that if you subscribe and sign in, you see fewer, or maybe even no, ads.
Time to install disco lights and TV in rooms (Score:2)
Reminds me of 2014 diffuse mirror research (Score:1)
Same same but different.
Creepy (Score:2)
Outside of some very specific situations, I don't see this being used by law enforcement, or multinational spies.
What it will be used for is companies like Amazon who will try to deep analyze what people are doing phy$ically and fine tune their telemetry data.
This is all kinds of ick, and stay the fuck away from my body!