Fast volumetric visual hull computation

Silhouette-based reconstruction techniques using multiple wide-baseline camera set-ups are popular building blocks to generate models of temporal human action. This contribution reviews visual hull computation algorithms. In particular we focus on the class of volumetric-based algorithms. We identify different processing strategies based on hierarchical, octree-based processing and suggest a new dynamic strategy that can significantly minimise the computational effort. The techniques discussed are evaluated with synthetic and real data sets. The savings demonstrated in computations on the real test sequences are in the order of a factor six on a 512×512×512 volumetric resolution.