A method of stabilizing a three-dimensional (3D) digital video image sequence includes generating a depth map for each stereoscopic pair of images using each of the image frames in a first-perspective (e.g., left) group and each of the image frames in a second-perspective group (e.g., right) and determining the background regions of each of the depth maps, defining the background regions in each of the image frames in one group among the first-perspective and second-perspective groups using the determined background regions, and calculating first global motion vectors between the consecutive image frames in the one group using only pixels of the defined second background regions.