Ohio prof develops CCTV people-tracker 'ware

[Eugen Leitl]

http://www.theregister.co.uk/2008/12/19/cctv_tracker_ware/

Ohio prof develops CCTV people-tracker 'ware

Paging Mr Orwell

By Lewis Page b" Get more from this author

Posted in Policing, 19th December 2008 10:22 GMT

Boffins in Ohio have taken another step towards the global surveillance
panopticon of the future, developing software which can autonomously track an
individual through a city using CCTV cameras.

James W Davis, associate prof at the Ohio State computer science and
engineering department, developed the new spyware with the aid of grad
student Karthik Sankaranarayanan.

Davis and Sankaranarayanan's code works by using a pan-tilt-zoom camera to
create a panoramic image of its entire field of view, and then linking each
ground pixel in the picture to a georeferenced location on a map. This means
that when the camera sees a person or vehicle, the computer also knows in
terms of map coordinates where it is looking.

That in turn makes it possible for a new camera to be trained on the target
as he/she/it passes out of the first one's field of view. In this way, a
subject can be followed automatically anywhere that the monitoring computer
has CCTV coverage. There's no need for a human operator to manually train
cameras around, using up man-hours and sooner or later making a mistake and
losing track.

"That's the advantage of linking all the cameras together in one system - you
could follow a person's trajectory seamlessly," says Davis.

For now, such camera networks are small and localised. However, the Home
Office here in the UK has said it would like to "create an effective cross
country strategic CCTV network". Such a network, combined with Davis and
Sankaranarayanan's new software, would allow plods or spooks to track people
completely hands-off. That said, until facial-recognition software gets a lot
better the computers would lose their target as soon as he or she left CCTV
coverage.

Not content with his efforts so far, Davis wants to go even further and write
code which can pick out people "engaging in nefarious behaviour".

"We are trying to automatically learn what typical activity patterns exist in
the monitored area, and then have the system look for atypical patterns that
may signal a person of interest," he says.

Such systems are already being trialled, and are known to be more than a bit
flaky. The panoramic-map software with its people-tracking abilities seems
more promising - from a surveillance operator's point of view, anyway.