The system reconstructs 3-D scenes from sequences of noisy binocular i
mages. First, the system establishes possible matches between the feat
ure pixels in the first image pair and assigns a confidence value to a
possible match. Each possible match is associated with a 3-D volume.
The possible matches are used to predict projections of associated 3-D
volumes onto the remaining image pairs. The new matches are used to u
pdate confidence values using a Kalman filtering related fusion algori
thm. Finally, matched pixels with high confidence values are used to r
econstruct a 3-D scene.