Temporal stereo disparity estimation with graph cuts

In this paper, we propose a temporal stereo disparity estimation method. Conventional stereo disparity estimation methods rely on matching costs regarding computation of intensity or position similarities. However, most applications do not consider the temporal dimension when estimating the disparity. In other words, previous approaches disregard potentially useful disparity information that is already estimated. Therefore, we exploit reasonable temporal disparity information for accurate disparity map estimation. First, we calculate the optical flow between two color frames. Then, the disparity from the previous frame becomes reliable information, shrinking the value of candidate disparity. Consequently, the proposed technique increases the temporal consistency of estimated disparity maps and outperforms per-frame methods when image noise is present.