Surveillance System Scans 36M Faces in One Second?

You are being watched. The government has a secret system, a machine that spies on you every hour of every day. I know because I built it. I designed the machine to detect acts of terror but it sees everything. Violent crimes involving ordinary people, people like you. Crimes the government considered irrelevant. They wouldn't act, so I decided I would. But I needed a partner, someone with the skills to intervene. Hunted by the authorities, we work in secret. You'll never find us, but victim or perpetrator, if your number's up... we'll find you.
 
MacGyver467 said:
You are being watched. The government has a secret system, a machine that spies on you every hour of every day. I know because I built it. I designed the machine to detect acts of terror but it sees everything. Violent crimes involving ordinary people, people like you. Crimes the government considered irrelevant. They wouldn't act, so I decided I would. But I needed a partner, someone with the skills to intervene. Hunted by the authorities, we work in secret. You'll never find us, but victim or perpetrator, if your number's up... we'll find you.
Click to expand...

Awesome, not only did you beat me to the punch, but I was going to cheat and paraphrase it. Now the question is, are YOU the person of interest?
 
Stonejaw said:
Its easy when all faces are exactly the same. LOL. =)
Click to expand...

Oooohhhh...I shouldn't be laughing...but I am :)

Reminds me of the story I read yesterday about how there are 200 CCTV cameras within one mile of George Orwells home. Oh George...if you could see us now.
 
No, not crazy. The comments in the video match what I expected - the stored images are being preprocessed sometime before the '1 second' match. Facial recognition technology is well developed both for detecting the location of faces in an image and categorizing their features.

It's that categorization that makes it so fast - you're reducing it from 300+KB of pixels to maybe 32 bytes of data representing the relationship of key points (distance of eyes from center, nose, etc). Each frame is preprocessed to detect faces and then create hashes of each face in the frame. To find a particular face among your stored security footage you run a match against that big list of face hashes. Close matches are then referenced back to the original video frame where a more detailed comparison can be done.

For comparison I've reached over 1 million image (not face) hash comparisons per second on desktop hardware. 36 million seems well within reach.
 
David_CAN said:
No, not crazy. The comments in the video match what I expected - the stored images are being preprocessed sometime before the '1 second' match. Facial recognition technology is well developed both for detecting the location of faces in an image and categorizing their features.

It's that categorization that makes it so fast - you're reducing it from 300+KB of pixels to maybe 32 bytes of data representing the relationship of key points (distance of eyes from center, nose, etc). Each frame is preprocessed to detect faces and then create hashes of each face in the frame. To find a particular face among your stored security footage you run a match against that big list of face hashes. Close matches are then referenced back to the original video frame where a more detailed comparison can be done.

For comparison I've reached over 1 million image (not face) hash comparisons per second on desktop hardware. 36 million seems well within reach.
Click to expand...

i was thinking the same thing.... comparing hashes is nbd
 
"You are being watched. The government has a secret system, a machine that spies on you every hour of every day."
 
MacGyver467 said:
You are being watched. The government has a secret system, a machine that spies on you every hour of every day.
Click to expand...
Sigh, fail on my part, I should have read the comments before I pasted mine above. But yes +1.
 
It can scan 36 million faces in one second.... provided you do lots of preprocessing on the pool of data to be scanned on the way into the system. That's not really that impressive algorithmically unless the pre-processing is really light weight.
 
raz-0 said:
It can scan 36 million faces in one second.... provided you do lots of preprocessing on the pool of data to be scanned on the way into the system. That's not really that impressive algorithmically unless the pre-processing is really light weight.
Click to expand...

I assume it's kinda like Shazam in a way. I'm going to guess it puts a digital fingerprint on each face for every frame it captures and then indexes it. When you select a face to search for, it just looks for that particular feature. Really fast index search (look what Google does in under a second) will give you the result.
 
I think we ought to make it easier for our people to spy on us. At birth we should all get large bar codes tattooed to our faces. Lets bump that 36M up to something more.

Kidding aside, that is scary tech. Impressive, but scary.
 
Something like...

montage1.jpg


Note, while many of the faces are the same person, they are different in pose.
 
David_CAN said:
No, not crazy. The comments in the video match what I expected - the stored images are being preprocessed sometime before the '1 second' match. Facial recognition technology is well developed both for detecting the location of faces in an image and categorizing their features.

It's that categorization that makes it so fast - you're reducing it from 300+KB of pixels to maybe 32 bytes of data representing the relationship of key points ... Close matches are then referenced back to the original video frame where a more detailed comparison can be done.
Click to expand...

What would be amazing is if he couldn't process 36M faces in a second.
But I don't think running a second more detailed comparison would make it in time, it's more likely that all key points are identified first, then a single bit pattern matching is applied.
From the screen sample, I also doubt 36M faces means 36M different persons, it looks more like he downloaded 36 pictures from 1 million Facebook users or something, so all these photographs are taken from different random angles and may increase his facial recognition score. A database with a single 2D picture of 36M individuals would not work as well.

And these all seem to be pictures from a HD digital camera, I doubt your basic streetcam has enough resolution for precise anthropometric measurements, you'll need a Enemy of the State 3D camera with rotating pictures. ^-^
 
MacGyver467 said:
You are being watched. The government has a secret system, a machine that spies on you every hour of every day. I know because I built it. I designed the machine to detect acts of terror but it sees everything. Violent crimes involving ordinary people, people like you. Crimes the government considered irrelevant. They wouldn't act, so I decided I would. But I needed a partner, someone with the skills to intervene. Hunted by the authorities, we work in secret. You'll never find us, but victim or perpetrator, if your number's up... we'll find you.
Click to expand...

Nice to see some other fans of Person of Interest. :cool:
 
Stonejaw said:
Its easy when all faces are exactly the same. LOL. =)
Click to expand...

I don't mean to be the buzzkill here, but it's technically HARDER to identify differences when there's a lot of similarity. It would be much easier to identify things vastly different than very similar.

...BUT, I did chuckle a bit :D
 
You must log in or register to reply here.
Back
Top