If you are orthogonal to the road, and you know the distance of the road part you see, it's obvious...
Otherwise, the general case is very difficult.
You have to estimate the motion of cars according to the point of view, maybe with size difference:
- extract the size at first point: a size of XXX means you are in point 1 of the road.
- extract size of the car at the end: a size of YYY < XXX or YYY > XXX (depending on your camera orientation) means you are in point 2 of the road.
- calculate time between two estimations
=> if you know the distance, you know the speed.
Step 1 & 2 aren't easy...
Good luck.