Multi-view Dense Depth Map Estimation

A novel dense depth map estimation algorithm is proposed in order to meet the requirements of N-view plus N-depth representation, which is one of the standardization efforts for the upcoming 3D display technologies. Hence, extraction of multiple depth maps is achieved from multi-view video. Starting from the piecewise planarity assumption of the scene, estimation of 3D structure of the patches, obtained through color-based over-segmentation, is achieved by plane- and angle-sweeping for every view independently. Markov Random Field (MRF) modeling is utilized for each view in pixel-wise manner in order to relax and refine the estimated planar models while incorporating visibility and consistency constraints. In this algorithm, the fusion of multiple depth maps is performed by updating belief values on the observed nodes based on depth and color consistency during the refinement step. The proposed method handles untextured surfaces, as well as depth discontinuities at object boundaries, due to its initial modeling of the scene as piecewise planar regions. The experimental results illustrate reliability and the robustness of the proposed algorithm for different type of scenes.