Knowing the direction and angle of the light source, one can calculate the approximate height that a point has.
However, the light direction and angle is usually unknown, which leads to poor quality images, or having to create multiple bump maps, each time trying different settings.
Light that comes from directly over the surface (behind the camera) makes it hard to distinguish the shadows.
Multiple light sources are also a problem, since they destroy important shadows required for the algorithms to work properly.
This process uses two images of one location taken from two separate lens on a camera, much in the same way humans do with their eyes.