research field in computer animation. Given a query motion clip, the
main task in content-based motion retrieval is to find all excerpts of a
motion capture database that are semantically similar to the query
motion clip. Facing the problem that semantically similar motions are
not necessarily numerically similar with respect to the angle and
3D-trajectories, we use relational motion features that show a high
degree of invariance to spatial deformations. To speed up the retrieval
process, we employ an index-based technique that uses designated frames
of the query motion, called keyframes, to quickly extract hit candidates
from the database. In particular, we introduce a new extraction
algorithm that handles time deformations of the keyframes, thus
accounting for temporal variations in semantically related motions. We
show that the runtime of our algorithm only depends linearly on the
length of the inverted lists of the keyframes, thus offering an
efficient supplement to computational more expensive techniques such as
dynamic time warping.