The epipolar plane image analysis method was introduced by Bolles [1] and later refined by Kim et.al [2], as a way to estimate depth from a sequence of images taken from different points lying along a straight line.
The method was initially aimed at situations where the disparities between successive images in the sequence were relatively large, but we want to apply this method to SkySat videos or to images issued from airborne hyperspectral sensors. In these cases disparities can be much smaller than one pixel, which may require some adaptations to the original method of Kim et al [2].
An implementation of this method is available and was successfully tested last year on SkySat videos, but several points can still be improved:
- The disparity confidence score can be improved
- One could use interpolation methods other than linear interpolation in order to better take into account the case of small disparities.
- One could augment the EPI with additional channels like in [3] to improve performance at coarse scales
The objective of this project is to understand [2] and its current implementation, and to implement one of the improvements listed above.
Supervision
Andrés Almansa, Gabriele Facciolo
Bibliographic References
[1] Bolles, R., Baker H., and Marimont D. Epipolar-plane image analysis: An approach to determining structure from motion. International Journal of Computer Vision, 1, 1987.
[2] Kim, C., Zimmer, H., Pritch, Y., Sorkine-Hornung, A., and Gross, M. Scene reconstruction from high spatio-angular resolution light fields. ACM Transactions on Graphics, 32(4), 1, 2013.
