Bundle adjustment algorithm

The bundle adjustment algorithm is a process whose aim is to take an initial estimate of the camera extrinsic parameters (their 3D positions and orientations), and an initial estimate of a set of 3D points in the take, and improves these estimates by a non-linear optimization process. The optimization refines both camera parameters and 3D point locations by finding the values that most accurately predict the locations of the observed corresponding 2D blob centers in the four camera images.

Bundle adjustment should generally give an improvement to the quality of the 3D point data (and camera positions). The most important time to perform bundle adjustment is during the 3D initialization process, after matching a set of facial seed point correspondences, and before performing automatic 3D initialization.

Bundle adjustment may also give a slight improvement to the overall results if performed once all tracking and editing and clean-up operations have been completed for a take.