A novel method for automatic 2D-to-3D video conversion

In this paper, we propose an efficient scheme to automatically convert existing 2D videos to 3D ones. The proposed method extracts motion information from two consecutive frames to estimate depth map for each of them. In the method, we first develop a region-based Graph cut method to fast and accurately perform motion segmentation, which is robust to large interframe motions. Then, a depth assigning step for the segments is conducted to obtain a smooth depth map for each frame. Experimental results on standard testing sequences demonstrate that our scheme achieves accurate motion segmentation and accordingly smooth depth map.