Graphics Reference
In-Depth Information
(a)
(b)
(c)
(d)
Figure 7.16.
Motion interpolation between a walking motion and a sneaking motion. (a) The
original walking motion. (b) The original sneaking motion. (c) Interpolation without dynamic
time warping. (d) Interpolation with dynamic time warping.
7.5.2
Motion Graphs
The previous section described how to interpolate between two given segments of
motion, but these segments must be specified by a user. This section describes how
to automate the problem of finding good transition points among a library of many
motion capture segments. Once we have this information, we can interactively string
together sub-segments of motion capture data that seamlessly satisfy a set of space-
time constraints (for example, that the skeleton be in a given pose at a given position
at a given time).
The underlying concept is called a
motion graph
[
254
], and is illustrated in
Figure
7.17
for a set of three motions. The idea is to determine a set of vertices
(time instants in each subsequence) and a set of edges (viable transitions within or
between subsequences) such that transitioning between vertices along edges results
in a natural-looking motion.
We already have the ingredients to create themotion graph; good transition points
can be identified directly by analyzing local minima of the pose distance function
c
θ
(
t
))
in Equation (
7.31
). Examples of such local minima are illustrated in
Figure
7.18
. We only accept minima whose distances are below a user-specified
threshold.
We can then compute the transition between every pair of identified transition
points using the interpolation methods of the previous section. Of course, minima
that are adjacent to each other within a single sequence don't need to be interpo-
lated since they're already connected by a sub-sequence of original motion capture
data. Each edge can be assigned a weight depending on the estimated quality of the
transition.
(
θ
(
t
)
,
