I believe it works something like this:
Case 1 - Camera flashes twice, takes two superimposed images. Someone measures the distance covered between flashes for each vehicle using white gridlines on the road. Distance divided by time etc.....
Case 2 - Unless culprit is photo'd/flashed twice, he's in the clear.
Case 3 - Same as 1.
Bikes are better placed than cars since they cannot be ID'd by cameras facing them.
Have a good w/e
VG