I don't know the specifics, but if I were to implement this I would start by constructing a signal map of an area, then comparing the signature of changes to potential routes.

By having a sufficiently accurate map, you could determine the likelihood of a route by the changes in signal. Even without pre-mapping there are some estimations that could be made: a signal that follows a clean sinusoidal curve with a certain rate of change could likely be modeled by rural highway driving.