Depth map estimation from motion for 2D to 3D conversion

The conversion of existing 2D videos to 3D videos is an important process for 3D video production. Generating an accurate depth map is the key starting pointfor this 2D to 3D conversion. This paper presents a novel method for generating the depth map of 2D image sequences from motion information. Different from the typical multi-view stereo method, this approach calculates the motion information directly from two sequential images which are captured by only a single moving camera. Classical constrained optical flow is known to be inaccurate for textureless regions and complicated areas. Our new approach uses mean shift segmentation algorithm and optical flow together to compute the depth map. The motion-based depth map and the segmented map are integrated into one depth map using breadth-first search methods. We find improved results compared to the optical flow method.