Incremental Poisson Surface Reconstruction for Large Scale Three-Dimensional Modeling

摘要

A novel Incremental Poisson Surface Reconstruction (IPSR) method based on point clouds and the adaptive octree is proposed in this paper. It solves two problems of the most popular Poisson Surface Reconstruction (PSR) method. First, the PSR is time and memory consuming when treating large scale scenes with millions of points. Second, the PSR can hardly handle the incremental reconstruction for scenes with newly arrived points, unless being restarted on all points. In our method, large scale point clouds are first partitioned into small neighboring blocks. By providing an octree node classification mechanism, the Poisson equation is reformulated with boundary constraints to achieve the seamless reconstruction between adjacent blocks. Solving the Poisson equation with boundary constraints, the indicator function is obtained and the surface mesh is extracted. Experiments on different types of datasets verify the effectiveness and the efficiency of our method.