From the above tutorial I've learned how to use Blender a bit better again :-)
In short here's what I did :
the original image is the texture of a plane with 'shadeless' material, disabled 'traceable' and 'shadow' shades and 'win' map input.
That plane is the child of the camera.
The scene is set to output at the same size as the original image.
The background image in blender is the original image.
A '0.5 alpha' - 'Z-transp' - 'only shadow' shadowplane is put according to the image's floor. It also serves as the floor in the blender 3D set up.
Because I've done something wrong, you can see the shadow of the backgroundplane on the street.
The balls I added to the image with the pink floyd ladies show how it's important to use the right lense for the camera.
If you want to add objects into a piece of movie, that also is explained in
this great tutorial.