Dense depth estimation

With the camera calibration given for all viewpoints of the sequence, we can proceed with methods developed for calibrated structure from motion algorithms. The feature tracking algorithm already delivers a sparse surface model based on distinct feature points. This however is not sufficient to reconstruct geometrically correct and visually pleasing surface models. This task is accomplished by a dense disparity matching that estimates correspondences from the grey level images directly by exploiting additional geometrical constraints.

This chapter is organized as follows. In a first section rectification is discussed. This makes it possible to use standard stereo matching techniques on image pairs. Stereo matching is discussed in a second section. Finally a multi-view approach that allows to integrate the results obtained from several pairs is presented.

- Image pair rectification

- Stereo matching

- Multi-view stereo

- Conclusion

Marc Pollefeys 2002-11-22