Animatable Human Body Reconstruction from 3D Scan Data using Templates

Many detailed 3D datasets of human bodies are available, and with current scanning technology, new ones are relatively easy to produce. As a result, more realistic looking animations become possible, but only after considerable processing: holes in the data, (too) high resolutions, and the laborious task of tting an animation skeleton to the data are the main problems to overcome. In this paper, we propose a reconstruction pipeline which solves these three problems by tting a template with surface and animation information to the scanned data. A correspondence between the scanned data and the template is set manually, with the aid of a tool that identies and visualises landmarks (characteristic points) on the mesh. Holes are handled using a ”vertex condence weight” scheme, with low weights assigned to vertices in or near holes (implying higher weights to template vertices, and vice versa). Initial tests on two detailed 3D human body scans produced highquality closed surface meshes, which are directly animatable.