Its a mix of pattern recognition and movement-prediction.

The pattern recognition (things like Neural Nets + Filters)
tries to identify a certain Object, and the
movement prediction tries to match this pattern consistently
to the same object, based on possible/predicted movement
when other objects (patterns) get close to it
(like a second car passing by).

It should be possible to do using optical and/or infrared data.

Its a classic military objective, for example to
track airplanes to be hit by a sidewinder rocked,
or track objects and identify them on a
radar screen.