Fast Adaptive Shape Matching Deformations

Abstract

We present a new shape-matching deformation model that allows for efficient handling of topological changes and dynamic adaptive selection of levels of detail. Similar to the recently presented Fast Lattice Shape Matching (FLSM), we compute the position of simulation nodes by convolution of rigid shape matching operators on many overlapping regions, but we rely instead on octree-based hierarchical sampling and an interval-based region definition. Our approach enjoys the efficiency and robustness of shape-matching deformation models, and the same algorithmic simplicity and linear cost as FLSM, but it eliminates its dense sampling requirements. Our method can handle adaptive spatial discretizations, allowing the simulation of more degrees of freedom in arbitrary regions of interest at little additional cost. The method is also versatile, as it can simulate elastic and plastic deformation, it can handle cuts interactively, and it reuses the underlying data structures for efficient handling of (self-)collisions. All this makes it especially useful for interactive applications such as videogames.