V. Scholz, T. Stich, M. Keckeisen, M.
Wacker and M. Magnor
Overview
We consider cloth motion capture as an important validation for and
alternative to simulation approaches. Real-time simulations require
simplifications which should be justified by real-world data. In many
cases the forces which drive cloth models are hard to model (e.g.
aerodynamics or friction forces from body contact). To close the gap
between model and experiment, an automatic measurement technique is
needed.
Our approach requires a custom-printed cloth pattern. We have chosen
M-arrays, a color code which encodes each point in the pattern by its
spatial neighborhood. Each 3x3 neighborhood of a point is unique and
can be used for point identification. Additionally, a triangle mesh for
the garment is constructed as input for the acquisition algorithm.
Based on photographs of the cloth panels, we design corresponding
triangle meshes and sew them together. The scene is recorded with eight
synchronized video cameras. Image feature recognition uses color
classification in HSV color space and edge detection for the ellipse
shapes. For matching the features to the cloth pattern we use a novel
region grow approach where seed points are identified by a 3x3 window
in the pattern that are grown using a search technique which adapts to
the pattern lattice structure. Using geometric camera calibration and
2D image feature positions, the 3D vertex coordinates are reconstructed
by triangulation. The fitted mesh is further processed by interpolative
hole filling.
