Meaningful Motion Blending
Marco da Silva, Eugene Hsu & Jovan Popivić
This project attempts to expand the size of a motion capture data set
by creating meaningful blends of existing motion clips. The space of possible
human motions is large and it is impractical to attempt to capture even
small regions of that space. Clever ways to compose existing data are
needed to maximize the utility of data that has already been collected.
Current techniques use simple cut and paste to compose motion clips of
body parts that don't conflict with each other. A motion composed in this
way may still look valid but throws away characteristics of the motions
that may add realism. A meaningful blending approach, however, would use
all available information to blend motions while maintaining any constraints
found in the input motions as well as semantic meaning. Preserving semantic
meaning will involve identifying which features of a motion give it meaning.
Given those features, this project seeks to solve the problem of blending
those features together in a meaningful way.
 Leslie Ikemoto and David A. Forsyth. Enriching a motion collection
by transplanting limbs. In Eurographics symposium on Computer Animation,
pp. 99-108, Grenoble, France, 2004.