A Comparison of Interpolation Methods for Virtual Character Upper Body Animation

Xingyu Lei, Purdue University

Abstract

The realistic animation of virtual characters can enhance user experience. Motion-editing methods such as keyframing and motion capture are effective for pre-determined animations but are incapable of real-time generation. Algorithm-based dynamic simulation and machine learning-based motion synthesis are procedural but too complex. This thesis explores an approach known as animation interpolation, which benefits from the strengths of both types of methods. Animation interpolation generates full animation sequences by assembling pre-defined motion primitives or key poses in real-time. The purpose of this thesis is to evaluate the naturalness of character animation in three common interpolation methods: linear Euler interpolation, spherical linear quaternion interpolation, and spherical spline quaternion interpolation. Many researchers have studied the mathematical equations, motion curves, and velocity graphs of these algorithms. This thesis focuses on the perceptual evaluation and the implementation of expressive upper body character animation. During the experimental studies, 97 participants watched 12 animation clips of a character performing four different upper body motions using three interpolation methods. The motions were based on McNeill’s classification of body gestures (beat gesture, deictic gesture, iconic gesture, and metaphoric gesture). After viewing each clip, the participants rated the naturalness on a 5-point Likert scale. The results showed that animations generated using spherical spline quaternion interpolation were perceived as significantly more natural than those generated from the other two interpolation methods.

Degree

M.Sc.

Advisors

Adamo, Purdue University.

Subject Area

Design|Applied Mathematics|Communication|Computer science|Mathematics|Robotics

COinS