alt Hacker NewsThe failure mode is incorrect detections. If the automated system says person 1 from camera 1 is the same as person 1 in camera 2, and then extrapolate out to 10,000 people and cameras, and then don't get it wrong. Because the BBC is never going to let you live it down if the person they arrested for stealing kinda sorta looks like the guy that actually did the stealing.
These are different problems, with different sensitivity/selectivity (and fail modes)
It's one thing to identify a person from a database, another from camera 1, then camera 2 etc. My guess is the latter has a much lower false positive rate