) Automatic correspondence analysis
The points can be matched from one image to the next by choosing matches which have the highest cross-correlation of image intensity for regions surrounding the points. The paths of the feature points are drawn in yellow here.
3) Outlier elimination
Due to misalignments or moving object in the scene some of the correspondences may be incorrect. To achieve a robust matching, a random sampling algorithm is employed to detect bad correspondences, called outliers.
4) Robust estimation of camera parameters
The camera parameters are estimated in an incremental fashion by using optimization technique applied to the good correspondences, called inlier.
5) Final refinement of the camera parameters
Finally, a refinement step is applied to all camera parameters of the sequence. This step tries to distribute the estimation error evenly over the sequence. Afterwards, virtual objects can be integrated into the real image sequence.