Global stereo matching formulations rely on piecewise constant disparity priors. For satellite images this is adequate for near-nadir views, where discontinuities are naturally associated to vertical walls. However, for slanted views this yields an asymmetry in the handling of occlusions where the “occluded wall” becomes slanted yielding to reconstruction artifacts. For this reason, rather than using an image grid, a formulation in terms of a geographic grid and altitudes is often preferred. For that the cost volume must be built with knowledge from the camera models. The main problems are in the
The objective of this project is to generalize SGM algorithm [1] presented in the course (or MGM [2]) to handle a terrain geometry cost-volume and compare it with the disparity based approach.
Supervision
Gabriele Facciolo, Carlo de Franchis, Enric Meinhardt
Bibliographic References
[1] H. Hirschmuller, “Stereo Processing by Semiglobal Matching and Mutual Information,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 30, no. 2, pp. 328-341, Feb. 2008.
[2] G. Facciolo, C. de Franchis, and E. Meinhardt. “MGM: A Significantly More Global Matching for Stereovision”, BMVC 2015.
