What's the conceptual difference between bundle adjustment and structure from motion?

Question

In my mind they both mean reconstructing 3D coordinates from matched points in 2D images. What's the difference between these concepts and multi-view stereo?

Which one do you call an algorithm that computes a sparse point cloud from keypoint matches, and requires both the cameras' external and internal parameters to be known a priori?

Answer 1

Structure-from-motion is the problem of recovering the 3D structure of the scene and the camera motion from a set of images. Bundle adjustment is a particular optimization algorithm used to solve it.

When the cameras' intrinsic parameters, and camera extrinsic parameters (i. e. camera poses) are known, you can actually compute the point cloud from the matching points using multi-view triangulation without bundle adjustment. You need to do non-linear optimization when your estimate of the camera poses is uncertain, and bundle adjustment is the standard algorithm used for that.