Abstract

Computers process and store human movement in a different manner from how humans perceive and observe human movement. The leading paradigm used by computer animation tools uses three tightly coupled models. These models, used in movies such as Toy Story, Final Fantasy: The Spirits Within, and Monsters Inc., are (1) a set of time signals, Q(t), that specify the kinematics of the movement, (2) a mapping, A, between Q(t) and the position, orientation, and posture of the human figure, and (3) a ``costume'' or ``visual appearance'' that specifies the outer appearance of the human body.

In sharp contrast to the exactness of computers, it is not well understood how we visually perceive human movements. It is believed that we utilize the motor control centers of our brains to recognize and interpret the movements of others. However, we do not know how observed movements are encoded or how they are translated into descriptions. Neither do we understand the process we use to translate descriptions or ``mental images'' of movements into physical movements.

In order to build higher-level computer animation tools for selecting, specifying, or modifying movements represented by computer models we need to know how the parameters of a movement, P(Q), affect our perceptions and judgements. We present a participant-based experimental and analytical methodology for gathering information on the relationships between three motion spaces: the first motion space is the ``mechanical motion space,'' a vector space of motion signals, Q(t); the second motion space is the ``psychological motion space'' in which humans encode and organize motions according to their features; and the third motion space is the ``linguistic motion space'' that humans use to describe movements using words.

We demonstrate our experimental and analytical methodology with two participant experiments that utilize computer animation displays of human walking figures to determine the effect of the parameters on judgements and descriptions of the movements. The first is a broad initial experiment to demonstrate the collection of judgements of the similarity of movements and descriptions of the movements from human observers using a wide range of human walking movements. The second is an in depth experiment to determine the properties of the psychological motion space by using a narrower range of walking movements that includes movements created by interpolating motion parameters. We conclude from the first experiment that the relationships between the motion spaces, for a small group of participants with similar backgrounds in social dance, are sufficiently explained by linear functions. We conclude from the second experiment that the psychological motion space does not have the metric properties necessary to treat similarity judgements as approximations of the distances between gaits in a metric space. We also conclude that the psychological motion space formed by similarity judgements is similar across a wide range of participant backgrounds (dancers, runners, and neither), genders (males and females) and the presentation of direction of walking across the screen (left to right or right to left).

Finally we suggest opportunities for future research and applications of our work in computer animation, human-computer interaction and psychophysics.

@PhdThesis{Harrison2001,
	author = {Jason Harrison, Ph.D},
	title = {Measuring and Comparing Human Walking Motions for Computer Animation},
	school = {UBC},
	year = {2001},
	supervisor = {Kellogg S. Booth and James J. Little},
}