How about the full-time job of one algorithm? Taken across N cameras, no actual person needs to watch any single camera. You can be targeted directly if the watcher controls a subset of N that captures you.
There is little difference between the output of one man and one camera following you around all day, and the result of N cameras stitched together algorithmically and watched by one person. In fact the machine managed version would be far superior given the increased range of capture N cameras (not required to be held by a human) provide.
Yes, the guy is annoying and I am surprised nobody beat him up. But the guy is not the point of this experiment - the camera is. An alternative experiment design could be placing the camera on a tripod at ground level and standing aloof a few feet away. I think just bringing the cameras down to ground level from the ceiling could be a big enough jump into the zone of discomfort for most people.
I'm not without my own concerns about surveillance, but let's not celebrate someone's antisocial behavior simply because he claims to be making a point. I think filming with a tripod, if it's not aimed at anyone in particular, would be far less offensive, but useless as a demonstration for that same reason.
I interviewed someone a few weeks ago whose prior job was maintaining such an algorithm for use in shopping malls. They had a database that you could issue queries like "Given that a person goes to store X, how long on average will it be until they return?" or "Are women who shop at Victoria's Secret likely to visit The Gap in the same month?" and so on. His job was applying that knowledge to affect what ads were displayed on the video screens that people walked past in the mall.
(Yes, the project was self-consciously inspired by "Minority Report". Apparently nobody found that creepy.)
Seems they are still around as: http://www.pathintelligence.com/
They were using USRP 1s (universal software radio peripheral, from Ettus Research) at the time, although they might have switched hardware platforms by now.
I've never had a chance to talk to them about false positives or negatives so I can't really put their numbers in perspective but they seem to project facial recognition as a 'solved problem.'
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.
Also claiming this is Anti-Social is the whole point, individuals can be called Anti Social because its easier for people to fight (be anti) them and make them conform. Governments and large corporations, on the other hand, are harder to fight and thus Society justifies allowing their behavior until enough people group to together to force a change.
This video is a metaphorical call to arms in that sense.