The objective of this project is to analyze and implement the method proposed in [1]. The proposed multi-view stereo method performs local fusion of disparity maps obtained with semi-global matching for reconstructing large scenes composed of hundreds of images. Careful modeling of the 3D errors allows to model the disparity uncertainty to improve the fusion.
Supervision
Gabriele Facciolo, Carlo de Franchis, Enric Meinhardt
Bibliographic References
[1] A Kuhn, H Hirschmüller, D Scharstein, H Mayer. A tv prior for high-quality scalable multi-view stereo reconstruction. International Journal of Computer Vision 124 (1), 2-17, 2017.