Bayesian surface reconstruction via iterative scan alignment to an optimized prototype

摘要: This paper introduces a novel technique for joint surface reconstruction and registration. Given set of roughly aligned noisy point clouds, it outputs noise-free watertight solid model. The basic idea the new is to reconstruct prototype at increasing resolution levels, according registration accuracy obtained so far, register all parts with this surface. We derive non-linear optimization problem from Bayesian formulation estimation problem. represented as partition unity implicit surface, which constructed piecewise quadratic functions defined on octree cells blended together using B-spline basis functions, allowing representation objects arbitrary topology high accuracy. apply standard data sets well especially challenging real-world cases. In practice, based reconstruction-registration algorithm avoids typical convergence problems in registering range scans substantially improves final output.