. Why does stereo matching on a rectified image pair only need a 1D search instead of a full 2D search?
Epipolar geometry for a rectified stereo pair guarantees the match lies on the same row, collapsing the search from 2D to a 1D scan.
. According to Z = f * B / d, what happens to estimated depth as disparity d increases?
Depth is inversely proportional to disparity: nearer surfaces shift more (larger d) and so have smaller Z.
. What does the left-right consistency check do in stereo matching?
A pixel that disagrees between the two matching directions by more than about a pixel is flagged as unreliable, which is especially effective at catching occlusions.
. What is the trade-off of increasing the block-matching window size in stereo matching?
A bigger window averages over more pixels (robust to noise) but mixes pixels from two different true depths near object edges (blurred boundaries); a smaller window has the opposite trade-off.