Fast Computing Adaptively Sampled Distance Field on GPU

In this paper we present an efficient method to compute the signed distance field for a large triangle mesh, which can run interactively with GPU accelerated. Restricted by absence of flexible pointer addressing on GPU, we design a novel multi-layer hash table to organize the voxel/triangle overlap pairs as two-tuples, such strategy provides an efficient way to store and access. Based on the general octree structure idea, a GPU-based octree structure is given to generate the sample points which are used to calculate the shortest distance to the triangle mesh. Classifying sample points into three types provides a well tradeoff between performance and precision, and when implementing the algorithm on GPU, these samples are also organized into blocks to share the triangles among threads to save bandwidth. Finally we demonstrate efficient calculation of the global signed distance field for some typical large triangle meshes with pseudo-normal method. Compared to previous work, our algorithm is quite fast in performance.