If the lens is reasonably close to planar over the area subtended by the car -- in other words, if it's not a fish-eye lens really close-up -- then the width of the car in the image is inversely proportional to the distance, i.e. width1 * distance2 = width2 * distance1.
Because you know both widths and distance1, you can determine distance2. The difference in distances, in metres, divided by the time in seconds, gives the speed.
If the lens is severely non-planar, you can always photograph yourself at various known distances and determine the exact diminution of image size as a function of distance. The factor by which image sizes change, will apply to the car as well as to you, and you can work out the distances and thus, again, the speed.
Edit: The road marking in your three example images are different, so either the camera has panned or zoomed. If it has panned, or if the images are all clipped from larger images shot at the same zoom, the analysis still applies. If the camera has zoomed, all bets are off.
Peter
Edited by peter (01/07/2005 09:14)