One very useful thing would be running facial recognition on all the faces the camera incidentally sees, in order to infer new connections between people.
I wonder how this works in practice - would they have high-magnification cameras raster-scanning the crowds and supplying the images to a facial recognition model?
Facial recognition is pretty much solid for most cameras now, though. Face registration, which Sony have been rolling into some E-mount cameras, would be more impressive.
There are ceiling-mounted facial recognition cameras for that now. Have a Customer who brought in some cameras for "people counting" from Xovis. Their's gear purported capabilities are down right scary.
Wow, that's cool! For years I've been wondering about all the places my face must have ended up with all the crowdshots people like to take at busy train stations and such; now, I see a way for this issue to become unnecessary as there could be cameras that just replace unnecessary faces. Paving the road to the future!
Gatwick airport had a ton of facial recognition cameras last time I was there.
They had a ring of leds around the lens that lights up in a rotating pattern, it really is difficult not to look directly into the lens due to that.
I’ve always dreamed of having a pair of glasses that would record everything that I see, and I could go back with that footage and use facial recognition to organize all the people that I met, and correlate that with other information that I have. It would definitely be an interesting improvement on my workflow at conferences and tradeshows.
To track someone through a crowd, you don't necessarily need continuous facial recognition. You can tag the identity to a person once using facial recognition or other means (credit card or ATM use if you have that data), and then revert to a more basic motion tracking system to follow them around.
I don't know much continuous object tracking across multiple cameras, but I highly doubt it's a difficult problem as long as the cameras have some overlap.
You add up to five faces, prioritized, and it focuses to them if it recognizes in the crowd of people.
reply