Spatiotemporal stereo is concerned with recovery of the 3D structure of a
dynamically changing scene from a sequence of stereo images. This paper
aims at computing temporally coherent disparity maps without explicit re-
covery of motion. We make use of a spatiotemporal volume paradigm and
consider extracted features we called stequels as basic matching primitives.
The stequel matching principle is developed. Extensive algorithmic evalu-
ation with ground truth data incorporated in both local and global corre-
spondence paradigms shows the great bene¯t of considering stequels as a
matching primitive that naturally incorporates local spatial and temporal
structure and its advantages in comparison to alternative ways of enforcing
temporal coherence in the stereo estimation procedure.
Mikhail Sizintsev, Richard P. Wildes