State Based Motion Transitions Using Framespace Interpolation

Most motion editing techniques require significant
computation resources or considerable manual
annotation. This thesis proposes analytical
classification and correspondence techniques to
support consistent interpolation between pairs or
quadruples of motion clips. The main contributions
are: a) Motion clips are labeled with locomotion and
generic kinematic joint level events; b) These event
labels are gathered into
states during inter-motion correspondence; c) An
efficient globally optimal correspondence between
states is performed, preserving state sequences; d)
Weights gathered from a spline function drive a
frequency and spatial domain motion transition,
preserving locomotion contact constraints. The thesis
extensively examines the application of the framework
to a wide variety of motions, unlike peer methods. It
proves the limitations of widely cited low-level
signal-correspondence techniques in addressing
transitions between different types of motions; e.g.
running and dancing. Useful additions include
unlimited motion chains and decoupled
upperbody-lowerbody interpolations. Key results and
comparisons have been posted at
http://www.comp.nus.edu.sg/~ashraf/framespace

Autorentext

Ashraf received his PhD in Computer Engineering fromNanyang Technological University, Singapore. He teaches GameDevelopment and Animation at School of Computing, NationalUniversity of Singapore. His research interests are in specialeffects, character animation, and computational aesthetics. Heenjoys drama, travel, photography, and guitaring.

Klappentext

Most motion editing techniques require significantcomputation resources or considerable manualannotation. This thesis proposes analyticalclassification and correspondence techniques tosupport consistent interpolation between pairs orquadruples of motion clips. The main contributionsare: a) Motion clips are labeled with locomotion andgeneric kinematic joint level events; b) These eventlabels are gathered intostates during inter-motion correspondence; c) Anefficient globally optimal correspondence betweenstates is performed, preserving state sequences; d)Weights gathered from a spline function drive afrequency and spatial domain motion transition,preserving locomotion contact constraints. The thesisextensively examines the application of the frameworkto a wide variety of motions, unlike peer methods. Itproves the limitations of widely cited low-levelsignal-correspondence techniques in addressingtransitions between different types of motions; e.g.running and dancing. Useful additions includeunlimited motion chains and decoupledupperbody-lowerbody interpolations. Key results andcomparisons have been posted athttp://www.comp.nus.edu.sg/~ashraf/framespace