10-14 Depth Estimation Method based on Likelihood of Block Matching among Multi-viewpoints

Abstract

In this paper, we propose a robust depth estimation method for multi-view video shot in an indoor environment. Our proposed method roughly reconstructs 3D model of a target object in 3D voxel space. Then, the method calculates likelihood of block matching for each voxel and decide real surface of an object. Experimental results show that our method improves the accuracy of depth estimation.