Generation of object-boundary-matched depth map is important to semi-automatic 2D to 3D conversion for high quality perception of converted 3D video. This paper proposes an accurate depth map generation method using an MRF-based initial contour tracking followed by a robust graph-cut-based contour refinement. The proposed MRF-based contour tracking puts a constraint on the variation of angular-radial distance for robust modeling of object motion in the presence of shape deformation. In addition, the proposed contour refinement eliminates ambiguous seeds, which might cause confusion in graph-cut segmentation, resulting in accurate contour in the presence of complex background. Experiment results show the superior tracking performance of the proposed scheme, especially when there is complex background and/or object deformation, indicating the power of the proposed scheme in object-boundary- matched accurate depth map generation.