Monday, April 18, 2011

Technology hype

An article, 'Brazilian Police Debut 'RoboCop' Glasses Ahead of World Cup', reports that the Brazilian police are testing "RoboCop" glasses...

It is claimed that "... the glasses are fitted with a tiny camera that scans up to 400 faces per second. It cross-checks those images against a database of criminals and terrorists, and flashes a small red light inside the glasses if a match comes up. Then the officer knows whom to home in on and whom to leave alone... On its optimal settings, the camera can scan 400 faces a second at a distance of up to 50 yards away. But the settings can be changed to recognize faces at a slower pace, at up to 12 miles away. Agostini said the camera and database compare 46,000 biometric points on a person's face, so the chances of mistaken identity are slim. "To the naked eye, two people may appear identical, but with 46,000 points compared, the data will not be beaten," he said..."

A YouTube video report also shows the system.

Ah, the eternal search for the science/technology silver bullet!

So, no doubt some cool technology that will work under certain, limited conditions. However, it seems grotesquely hyped. Let's think through some of this... Scanning faces at the rate of 400 per second the glasses will flash a red light to let the wearer know that a match has been made. Tough for the wearer to react and recognize the "perp." OK, so perhaps it stops when it finds a match... good idea? What if there are 20 matches? And of course it's only as good as the database against which the matches are being made...

The above assumed that the details provided were correct, which itself seems like a stretch. A camera that size that can distinguish (at 150 feet) between two people that appear identical to 'the naked eye'? Bah! That can also recognize people (albeit at a slower rate) at a range of 12 miles? Yeah, right! Look at the video feed in the YouTube link (two pictures abstracted below), does it seem likely that the faces of the people in the distance can be reliably scanned (and they are not all that far away)?

OK, so the YouTube demo/report appears somewhat more realistic than the article would appear to be. Note that the policeman with the scanner is standing fairly still, and also appears to be scanning people who are not all that far away, and who are also being funneled using metal barriers... And that there's a lot of equipment in addition to the on-glasses component!

Certainly, this technology probably does work under these optimal conditions, but the article and reporting seem terribly hyped! Finally, given all the other cameras shown (fixed on the building, movable on the police van and police helicopter), why wouldn't you hook up the technology to those and identify and intercept any 'problems' further away from the venue, versus as they walk up to the check point?

No doubt once the "RoboCop" glasses flag an individual, they can be interrogated, using PCASS to definitively (ahem!) determine their truthfulness (cough!).

A last item, any report that ignores any mention of accuracy measures (e.g. the False Rejection Rate, or Type 1 Error, in which a valid test subject is incorrectly rejected, and the False Acceptance Rate or Type 2 Error in which the test subject is incorrectly accepted) is automatically suspect as far as this blogger is concerned....

No comments:

Post a Comment