论文信息 - A simple and efficient way to compute depth maps for multi-view videos

A simple and efficient way to compute depth maps for multi-view videos

This paper deals with depth maps extraction from multi-view video. Contrary to standard stereo matching-based approaches, depth maps are computed here using optical flow estimations between consecutive views. We compare our approach with the one proposed in the Depth Estimation Reference Software (DERS) for normalization purposes in the ISO-MPEG 3DV group. Experiments conducted on sequences provided to the normalization community show that the presented method provides high quality depth maps in terms of depth fidelity and virtual views synthesis. Moreover, being implemented on the GPU, it is far faster than the DERS.

Gael Sourimant

[1] Daniel Cremers,et al. Anisotropic Huber-L1 Optical Flow , 2009, BMVC.

[2] Berthold K. P. Horn,et al. Determining Optical Flow , 1981, Other Conferences.

[3] D. Scharstein,et al. A Taxonomy and Evaluation of Dense Two-Frame Stereo Correspondence Algorithms , 2001, Proceedings IEEE Workshop on Stereo and Multi-Baseline Vision (SMBV 2001).

[4] Peter J. Huber,et al. Robust Statistics , 2005, Wiley Series in Probability and Statistics.

[5] Moshe Porat,et al. Correlation-based approach to color image compression , 2007, Signal Process. Image Commun..

[6] Horst Bischof,et al. A Duality Based Approach for Realtime TV-L1 Optical Flow , 2007, DAGM-Symposium.

[7] Thomas Brox,et al. Universität Des Saarlandes Fachrichtung 6.1 – Mathematik Highly Accurate Optic Flow Computation with Theoretically Justified Warping Highly Accurate Optic Flow Computation with Theoretically Justified Warping , 2022 .