The Lucas-Kanade sparse optical flow method performs a least-squares fit on the motion of a 3x3 pixel neighborhood (9 points) around each target point. The algorithm assumes small movement between consecutive frames to satisfy the brightness constancy constraint.