A Probabalistic Framework for Space Carving

Abstract

This paper introduces a new probabilistic framework for Space Carving (cite Kutulakos99). In this framework each voxel is assigned a probability, which is computed by comparing the likelihoods for the voxel existing and not existing.

This new framework avoids many of the difficulties associated with the original Space Carving algorithm. Specifically, it does not need a global threshold parameter, and it guarantees that no holes will be carved in the model. This paper also proposes that a voxel-based thick texture is a realistic and efficient representation for scenes which contain dominant planes. The algorithm is tested using both real and synthetic data, and both qualitative and quantitative results are presented.

see also Publications Page and Web Presentation (ICCV2001) and Data sets

Results

The Master's Lodge image sequence

Trinity College Chapel sequence

The Great Court fountain

Tennyson - Trinity College

Tennyson - Background removed

Pollefeys castle sequence