Non-rigid 3D shape recovery using linear stereo factorization
We propose a method to compute reliable 3D models of deformable structure from stereo images. Our experiments with real data show that by merging the information of the stereo pair we can achieve improved reconstructions. Note that our focus is on the recovery of flexible 3D shape rather than on the correspondence problem.
 |
The stereo algorithm needs reliable tracks from the stereo pair sequences. A video (avi, 5.03MB) shows the subject face performing small rigid motion and face deformations. |
 |
Shape reconstruction obtained from the factorization algorithm applied only to the right camera tracks. A video (avi, 2.81MB) shows a wrong shape reconstruction. The monocular factorization is not able to estimate a correct depth if the general rigid motion is small. |
 |
Shape reconstruction obtained from the factorization algorithm applied only to the left camera tracks. A video (avi, 2.92MB) shows again a wrong shape reconstruction. |
 |
Shape reconstruction obtained from the stereo factorisation applied to the stereo camera tracks. A video (avi, 3.14MB) presents major improvements leading to a correct shape estimate. |
Reference on Linear Stereo-Motion:
-
A. Del Bue and L. Agapito, "Non-rigid 3D shape recovery using stereo factorization," Asian Conference of Computer Vision, vol. 1, pp. 25-30, 2004.
@Article{DelBue:Agapito:2004,
author = {A. {Del Bue} and L. Agapito},
title = {Non-rigid {3D} shape recovery using stereo factorization},
journal = {Asian Conference of Computer Vision},
year = {2004},
month = {January},
volume = {1},
address = {Jeju, South Korea},
pages = {25-30}
}