Supplementing Frequency Domain Interpolation Methods for Character Animation

The animation of human characters entails difficulties exceeding those met simulating objects, machines or plants. A person's gait is a product of nature affected by mood and physical condition. Small deviations from natural movement are perceived with ease by an unforgiving audience. Motion capture technology is frequently employed to record human movement. Subsequent playback on a skeleton underlying the character being animated conveys many of the subtleties of the original motion. Played-back recordings are of limited value, however, when integration in a virtual environment requires movements beyond those in the motion library, creating a need for the synthesis of new motion from pre-recorded sequences. An existing approach involves interpolation between motions in the frequency domain, with a blending space defined by a triangle network whose vertices represent input motions. It is this branch of character animation which is supplemented by the methods presented in this thesis, with work undertaken in three distinct areas. The first is a streamlined approach to previous work. It provides benefits including an efficiency gain in certain contexts, and a very different perspective on triangle network construction in which they become adjustable and intuitive user-interface devices with an increased flexibility allowing a greater range of motions to be blended than was possible with previous networks. Interpolation-based synthesis can never exhibit the same motion variety as can animation methods based on the playback of rearranged frame sequences. Limitations such as this were addressed by the second phase of work, with the creation of hybrid networks. These novel structures use properties of frequency domain triangle blending networks to seamlessly integrate playback-based animation within them. The third area focussed on was distortion found in both frequency- and time-domain blending. A new technique, single-source harmonic switching, was devised which greatly reduces it, and adds to the benefits of blending in the frequency domain.

Name dc:type.qualificationname

phd

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of East Anglia

Year dc:date.issued

2012

Author dc:creator

• Molnos, MRL

Language dc:language

en