US military tracker-droids to 'consider humans as fluid'
Traffic algorithm upgrade for airborne spyeyes
Pentagon propellerhead chiefs have hatched another sinister surveillance scheme. The plan is to add "flow based" theories of urban traffic movement - which assume that humans en masse behave like fluid moving through a network of pipes or channels - to conventional radar/camera based tracking of vehicles.
One need hardly say that the body behind the Flow-based Information Theory Tracking (FITT) scheme is none other than DARPA, the Defense Advanced Research Projects Agency. When you're a hammer, all the problems start to look like nails; when you're DARPA, all the problems start to look like insignificant fleshies scurrying like cockroaches among their inefficient human warrens as your automated aerial robo-surveillance network peers down god-like from above.
The DARPA boffinry chiefs explain their aspirations to enhance their aerial spy networks thus:
The FITT program assumes that a sensor (e.g., radar or a set of cameras) observes a ground area...
As ground target densities increase in more urban areas, existing trackers often lose target track due to nearby confusers, and operate with limited hypothesis depths to avoid computational overload. In addition, urban traffic offers many constraints that conventional trackers do not exploit. For example, ground vehicles cannot pass thru each other, they cannot go beyond the bounds of typical urban roads, and they generally obey cultural conventions and traffic laws. Under these constraints, ground traffic behaves somewhat like a fluid, and the FITT Program expects to develop new tracking algorithms based on this fluidic viewpoint.
Thus, even if all the radar blips representing moving cars etc. start to run together, or if a car could no longer be seen visually due to an intervening building, lorry etc., a FITT-enhanced tracker system would still cope. It would be able to follow the position of a target vehicle at least for a while by monitoring the traffic flowing along.
Likewise, a FITT algorithm might be able to infer behaviour. If it was watching a vehicle approaching a T junction, but couldn't observe the junction itself, it might note the vehicle slowing down and work out that it planned to turn at the junction rather than carrying straight on. Etc etc.
The initial DARPA workshop for people who'd like to design such kit takes place in a couple of weeks, and will among other delights feature a probably very interesting "Tracking State-of-the-Art" presentation. Full details from DARPA here in pdf. Strangely, it doesn't seem to be classified top secret, as you might expect. ®
Sponsored: Benefits from the lessons learned in HPC