In this paper, Ho and Chung describe an algorithm that obtains environmental depth information from a pair of calibrated affine cameras that move as a rigid unit through a static environment. They demonstrate that if robust temporal correspondences are obtained from each camera, the stereo correspondence problem can be solved trivially--provided that temporal correspondences can be found that do not violate certain geometric constraints. Experiments with synthetic and real imagery are presented.
The approach described in this paper seems quite robust provided that good temporal correspondences can be established in each camera over the entire sequence. For many objects and camera motions, this may not be practical, and the extension of the approach to deal with these sorts of conditions would make the approach more applicable for real-world applications.