The generalizing power of deep convolutional neural nets is much greater, but (1) it doesn't really matter if you don't have anything similar in your training dataset, and (0) CNNs are used for face _recognition_, and for it you still have to detect face first. And in detection, VJ is still king.
Granted my domain of knowledge is more aligned with the recognition side, i'm pretty sure Viola-Jones has been replaced as 'king' with more robust approaches to detection. Here is a paper published just last year in CVPR:
I've been working on an idea for a while, autonomous police cars. They'll spot a criminal, pick them up and drive them to jail. It could save billions!
Sounds like my stupid (brilliant?) idea, an autonomous car lift. Travel under/around speeding or otherwise criminal drivers, travel at the same speed and pick the vehicle up (from the frame, not the wheels).
"10:01 AM", by Alexander Malec, 1966. Same idea, but for flying cars.
Reckless driver kills child pedestrian. Gets caught, grabbed in air by large police vehicle. Driver has disabled vehicle recorder. Backup vehicle recorder hidden in lug nut is played back.
It is just a university research group presenting a development prototype for a police car - so it's not likely something that will be on the roads anytime soon (though the Hungarian article speculates that this would "come in handy for the september G20 meeting in Tsengdu and could be introduced there").
Their prototype can "automatically scan all faces in a sixty meter circle around the car" and be able to reconcile it with lists of suspects, even at speeds of 120km/h. It is also able to recognize and store all number plates visible around the cop car.
The article doesn't mention it directly, but it seems it will scan actively.
Thanks for the info, the article says that the creating of a similar system is also in progress in the United Arab Emirates.
Surprised this isn't already a thing in production. Police cars have had automatic license plate detection coupled to databases of known stolen cars etc. for many years. Likewise cars have had pedestrian detection cameras for a few years.
All the technical bits are there, and I can't imagine it would be a difficult issue legislation/privacy-wise.
> Szép világ lenne, de erre minden bizonnyal még több évtizedet kell várnunk.
> (A beautiful world it would be, but we will have to wait decades.)
It seems that the current generation of Hungarian journalists has forgotten the lessons of overzealous policing during the bad old days from the late 1930s to the early 1990s.
Seems like it, the captured photos (in 360 degrees) will be compared within milliseconds with the list of the wanted criminals in the central database. Even when the car is moving with 120 kmph.
does the Chinese government really care? I ask because this is the type of thing that only enters a conversation when 'effectiveness'/'accuracy' becomes a discussion down the line. I doubt that China gives a s* frankly, which makes this even scarier...the data used to show it works will like be bad by design.
It's a quantum leap in surveillance tech -- the latency for nation-wide tracking of individuals will be phenomenally reduced. HUMINT resources can be more effectively allocated and target acquisition times can be reduced.
That said, the false positive rate is guaranteed to be high at first but the tech will be improved, furthermore the data will be correlated with other's rather than sole-sourced.
Sure they won't mind if there's collateral damage, but it's useless if it wastes time and effort by leading to the wrong person too often.
Really important question, but unfortunately the article doesn't mentions it. What do mention is that for a successful identification only the three-quarter of the face is needed.
We published a paper on this a few years ago! Short answer: algorithms exhibit different accuracies on different races, genders, and ages.
B. F. Klare, M. J. Burge, J. C. Klontz, R. W. Vorder Bruegge and A. K. Jain, "Face Recognition Performance: Role of Demographic Information," in IEEE Transactions on Information Forensics and Security, vol. 7, no. 6, pp. 1789-1801, Dec. 2012.
Computers certainly can be racist when it comes to visual recognition tasks. Different races do look different, after all. For example, there have been several cases of face tracking software failing to detect black people.
In the case I think you're referencing, they were detected, just not as people.
It's a bit of a cop-out just to say, "shucks, computers do the darnedest things!" Humans coded, tested, and deployed that service. They let something fall through the cracks, the awful implications of which detract quite a bit from all the good the service might ever accomplish.
There are a bunch of different examples out there, just search for "face detection black people." In one example, Google Photos tagged a pair of unfortunate folks as gorillas. In another, HP had some webcams that tracked faces but couldn't track black people. Kinect had a similar problem. Nikon had a "did someone blink?" message that produced a lot of false positives with Asian people.
My point isn't to excuse it, and indeed I'm not even trying to address the moral dimensions of it. My point is just that if this is a Chinese product developed for use in China, it's entirely possible that it won't work well on non-Asian faces without additional work.
But surely, if you're a criminal, then you'll probably take measures to ensure your face doesn't get scanned/recognized? So this is either a measure to intimidate law-abiding citizens into staying docile, or just a corruption scheme to appropriate the funds from the budget (subcontract the development of this technology to a corporation run by one of your friends).
https://cvdazzle.com/