Ask HN: What's the name of this method/agorithm?

Say I have a video from a random camera somewhere in the world. The camera view is fixed and we don’t know neither its intrinsic nor extrinsic parameters.

Is there a way of automatically place 3D objects in the scene? That is automatically find a ground plane and somehow calibrate it using reference objects.

Is this scalable? Say I have thousands of different videos, now any manual step is not feasible. Is there a way of using the video itself for this calibration problem?