Copyright is held by the author / owner(s). SIGGRAPH Asia 2010, Seoul, South Korea, December 15 – 18, 2010. ISBN 978-1-4503-0439-9/10/0012
Perceptual Evaluation of Human Animation Timewarping
Martin Prazˇa´k Rachel McDonnell Carol O’Sullivan
Graphics, Vision and Visualisation Group, Trinity College Dublin
-1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6
Motion speed change [m/s]
10
0
50
0Ti
m
ew
ar
p
de
te
ct
io
n
ac
cu
ra
cy
[%
]
(a) (b) (c)
Figure 1: (a) screenshot from running experiment (side view, stick figure); (b) stick figure and geometrical model used in the study; (c)
experiment results in absolute differences in speed of the locomotion, showing that speeding up the motion produces severe perceptual
artifacts while event significant slow down is perceptually acceptable.
Abstract
Understanding the perception of humanoid character motion can
provide insights that will enable realism, accuracy, computational
cost and data storage space to be optimally balanced. In this sketch
we describe a preliminary perceptual evaluation of human motion
timewarping, a common editing method for motion capture data.
During the experiment, participants were shown pairs of walking
motion clips, both timewarped and at their original speed, and asked
to identify the real animation. We found a statistically significant
difference between speeding up and slowing down, which shows
that displaying clips at higher speeds produces obvious artifacts,
whereas even significant reductions in speed were perceptually ac-
ceptable.
1 Computer Animation and Perception
Computer animation and motion perception are closely related
fields, as the result of motion synthesis is always presented to a live
observer. In this study, we focus on motion timewarping, a method
used in both parametric models and state machines for matching
and transitioning between clips of the same type with different tim-
ing or speed. Timewarping, and particularly dynamic timewarping,
originates in speech recognition and was succesfully used, along-
side other signal processing techniques, on animation data (e.g.,
Bruderlin and Williams [1995]). However the perceptual implica-
tions of such manipulations have not been extensively studied and
are hard to predict. Our evaluation approach is closely related to
the work of Reitsma and Pollard [2003], but we focus on human
locomotions rather than generic ballistic motion.
2 Experiment Design
Five motion captured clips of a walking animation served as the
stimuli for our experiment. These five animation speeds covered a
normal range of human walking, ranging from 0.8 m/s to 2.4 m/s
with 0.4 m/s increments. For each clip, we created 4 other versions
using timewarping to match the speed of the other clips, leading to
a total of 25 clips (e.g., the 1.2 m/s motion was slowed down to
0.8 m/s and speeded up to 1.6, 2.0 and 2.4 m/s). We hypothesised
that timewarping would be less noticeable if the timewarped speed
is close to the original speed of the clip.
The experiment consisted of sequences depicting two animated
characters side-by-side (Figure 1). Both characters (Figure 1b)
were either stick figures or geometric models (the model’s realism
was found to affect perceptual sensitivity to errors in motion [Hod-
gins et al. 1998]), facing forward or sideways (to test if motion error
sensitivity is affected by viewpoint), with each simultaneously dis-
played pair using the same setup. One character’s animation con-
sisted of the original motion, randomly placed on the left or right
side of the screen, while the other was timewarped to match its
speed. We also tested the real animation against itself as a control
case.
Sixty naive participants from the general public (64% male and
36% female) volunteered for this experiment. The instruction sheet
indicated that one of the motions was a real captured motion and
the other one was synthetically edited. The task was to indicate
which of the two animations was the real motion by clicking the
left or right mouse button. Each participant completed 100 trials in
randomized order (25 motion clips, 2 models, 2 viewpoints).
3 Results
Using a 3-factor repeated measures ANalysis Of VAriance
(ANOVA) with factors model type (stick or geometric), viewpoint
(front or side), and speed (speed up or slow down), we found a main
effect of model type (F1;59 = 8:09; p < 0:007). Post-hoc analy-
sis using Newman-Keuls comparison of means showed that partici-
pants were more sensitive to timewarping on the geometric models
than on the stick-figure. This implies that the effect of timewarp-
ing is more noticeable on more realistic characters, consistent with
previous results [Hodgins et al. 1998]. No main effect of viewpoint
was found, which implies that side and front views did not affect
sensitivity to timewarping artifacts.
A main effect of speed was found (F1;59 = 133; p < 0:000001),
where participants were able to notice speeded up motions on av-
erage 80% of the time vs. 47% of the time for slowed down an-
imations. These results were consistent across conditions. Since
we used a 2AFC (2 alternative forced choice) paradigm, this sug-
gests that participants were simply guessing when they viewed a
slowed down animation beside a real one. Therefore, consistent