Disparity map using suboptimal cost with dynamic programming

Abstract

1D optimization methods based on dynamic programming (DP) stereo are of practical interest because it can reconstruct an observed 3D optical surface very quickly and thus has potential for real-time applications. While being efficient, its performance is far from the state of the art because the vertical consistency between the scanlines is not enforced. 1D optimization based on dynamic programming for stereo correspondence is re-examined by applying it to the vertical consistency between the scanlines as opposed to the individual scanlines. To do this, a pixel is allowed to have a disparity with possibly sub-optimal cost for it in two directions. Thus, the proposed algorithm is a truly global optimization method because disparity estimate at one pixel depends on the disparity estimates at all the other pixels, unlike the scanline based methods. Proposed algorithm is evaluated on the benchmark Middlebury database. The algorithm is very simple, so the proposed algorithm should be a good candidate for real time implementation. The results are considerably better than that of the scanline based methods. While the results are not the state of the art, the proposed algorithm offers a good trade off in terms of accuracy and computational efficiency.