An automatic framework for quadrilateral surface reconstruction with partitions from 3D point clouds

Currently, three-dimensional (3D) point clouds are widely used in the gaming and film industries. Inspired by the reverse process of polycube-based parametrical mapping, we present an automatic framework that can directly reconstruct a quadrilateral surface with part-based partitions from 3D point clouds and restore their intricate details while remaining resistant to noise. This innovative framework holds potential applications in various fields, including the enhancement of skeleton and rigging processes in computer animation. In our framework, we initially generate a one-dimensional medial skeleton from 3D point clouds. This skeleton guides us in determining the direction mutation and area mutation of 3D point clouds, aiding in the partitioning of the initial quadrilateral surface into meaningful parts. Following this, we evolve our initial quadrilateral surface by moving each vertex toward its normal direction, which is bounded by an unsigned distance field derived from the 3D point clouds. Our framework ultimately generates a high-quality quadrilateral surface that recovers fine details of 3D point clouds with part-based partitions that reflect meaningful partitions of the original object.