next up previous contents
Next: Lightfield modeling and rendering Up: Lightfield model Previous: Lightfield model   Contents

structure and motion

To do a dense lightfield modeling as described below, we need many views from a scene from many directions. For this, we can record an extended image sequence moving the camera in a zigzag like manner. The camera can cross its own moving path several times or at least gets close to it. Known calibration methods usually only consider the neighborhoods within the image stream. Typically no linking is done between views whose position is close to each other in 3-D space but which have a large distance in the sequence. To deal with this problem, we therefore exploit the 2-D topology of the camera viewpoints to further stabilize the calibration. We process not only the next sequential image but search for those images in the stream that are nearest in the topology to the current viewpoint. Typically we can establish a reliable matching to 3-4 neighboring images which improves the calibration considerably. The details were described in Section 5.2.2. We will also show how to use local depth maps for improving rendering results. To this end dense correspondence maps are computed for adjacent image pairs of the sequence (see Chapter 7).


next up previous contents
Next: Lightfield modeling and rendering Up: Lightfield model Previous: Lightfield model   Contents
Marc Pollefeys 2002-11-22