Global Depth from Epipolar Volumes--A General Framework for Reconstructing Non-Lambertian Surfaces

Citations Over TimeTop 20% of 2006 papers

Abstract

Using epipolar image analysis in the context of the correspondence finding problem in depth reconstruction has several advantages. One is the elegant incorporation of prior knowledge about the scene or the surface reflection properties into the reconstruction process. The proposed framework in conjunction with graph cut optimization is able to reconstruct also highly specular surfaces. The use of prior knowledge and multiple images opens new ways to reconstruct surfaces and scenes impossible or error prone with previous methods. Another advantage is improved occlusion handling. Pixels that are partly occluded contribute to the reconstruction results. The proposed shifting of some of the computation to graphics hardware (GPU) results in a significant speed improvement compared to pure CPU-based implementations.

Related Papers

• → Epipolar geometry estimation via RANSAC benefits from the oriented epipolar constraint(2004)34 cited
• → Epipolar Spaces for Active Binocular Vision Systems(2007)5 cited
• The estimation algorithm of epipolar geometry based on sub-pixel match strategy(2010)
• A Robust Correspondence Using the Epipolar Geometry from Two Un-calibrated Images(2006)