Depthless Streaming of Depth-based 3d Videos

In this brief on-going research paper, we summarize our current work on reconstructing the depth map from a fusion of multiple estimated depth maps that are generated from a number of multiple monocular cues. We first analyze a ground truth depth map to extract a set of depth cues or statistics. Then, using these depth cues, we process the colored reference video and generate an estimate of the depth map per monocular cue. The monocular cues we consider in this work are luminance, chrominance, motion, and texture. Finally, we plan to integrate all these estimates into one final depth map. We tested our approach on different video sequences with different monocular properties. The results show that the extracted depth maps generate a 3D video with quality close to the video rendered using the ground truth depth map. The ultimate goal of this research is to stream the reference video with very few side depth information for each frame and thus eliminate the need for the depth streaming channel in such systems.