Efficient Voxel Marking for Hierarchical Volumetric Fusion

Abstract

When fusing depth images into a 3D volumetric model, a crucial task is to mark macro-cells as empty or as intersected by the noisy surface represented by the depth image. This paper proposes a simple marking algorithm for the GPU implementation of hierarchical volumetric fusion. The method is based on multi-level DDA ray-casting. The GPU implementation is of scattering type, but we also show a solution to avoid atomic writes, which improves performance.