Regarding background sources, the fact that the target of interest is moving could actually be a great benefit for background rejection. Place two detectors on the path separated by some distance, knowing how long it takes for a typical vehicle to travel between them, and you can filter out simultaneous spikes as background fluctuation. Modern photon counting detectors are pretty amazing, and it would take only a small number of temporally correlated discrete events at each detector to reach the probability threshold of alarm.

I don't work on these systems, obviously - those who do aren't talking - and this is just idle speculation. But I do know that lots of DHS money is shoveled into R&D on this kind of thing every year and I would not be surprised if some of the tech does work.