Grafting Locomotive Motion
Abstract
The notion of transplanting limbs to enhance a motion capture database is
appealing and has been recently introduced. A key difficulty in the process is
identifying believable combinations. Not all transplantations are successful;
we also need to identify appropriate frames in the different clips that are
“cutpasted.” In this paper, we describe motion grafting, a method to synthesize
new believable motion using existing motion captured data. In our deterministic
scheme designed for locomotive actions, motion grafts increase the number of
combinations by mixing independent kinematics chains with a base motion in a
given clip.
Our scheme uses a cluster graph data structure to establish correlation among
grafts so that the result is believable and synchronized.
The Proposal
We address the problem of increasing the motion repertoire of a given motion
database. We achieve this by synthesizing new motion as a composition
of sub-hierarchy motion from two different clips. The primary
challenge for this task is the detection and accounting of cross hierarchy
correlation. We use foot plant information as a synchronization signal to
implicitly handle the correlation for the class of regular locomotive motions
walk, run, jump and stand.
Grafting frame work
-
Annotate motions in the motion database with footplant constraint information
-
Select regular locomotive motion stand, walk, run, jump
-
Build cluster graphs for upper and lower body hierarchies of the selected
motion clips
-
Select a pair of clips for grafting using the upper body cluster graph. The
clips selected should follow the composition class rules specified below. The
upper body motion of the two clips should have some cluster graph node in
common i.e. there exists a transition between the two upper body actions.
-
Determine foot plant frames near the transition frame by searching the lower
body cluster graph.
-
Resample the graft clip to match the sampling rate of the target clip, again
using the distance between corresponding footplants as the resampling metric
-
Synthesize the new clip.
Foot plant detection
We detect footplants by locating frames where
-
Hell or ball is close to ground.
-
Relative displacement of either heel or ball is small (less than a specified
threshold).
Figure shows the heel and ball positions at frames determined to be planted on
the ground.
Our contribution
Results
Figure on the left show the exaggerated swinging arm movement of the middle clip,
grafted on to the lower body motion of the top clip. Figure on the right shows, a
basketball dribble motion being grafted on to the lower body motion of the top
clip.
Downloads
