SIGGRAPH Asia 2015
Generalizing Wave Gestures from Sparse Examples for Real-time Character Control

Helge Rhodin1 James Tompkin2 Kwang In Kim3 Edilson de Aguiar4 Hanspeter Pfister2 Hans-Peter Seidel1 Christian Theobalt1
  1MPI für Informatik 2Harvard Paulson SEAS 3Lancaster University 4Federal University of Espirito Santo


Motion-tracked real-time character control is important for games and VR, but current solutions are limited: retargeting is hard for non-human characters, with locomotion bound to the sensing volume; and pose mappings are ambiguous and not robust with consumer trackers, with dynamic motion properties unwieldy. We robustly estimate wave properties — amplitude, frequency, and phase — for a set of interactively-defined gestures, by mapping user motions to a low-dimensional independent representation. The mapping both separates simultaneous or intersecting gestures, and extrapolates gesture variations from single training examples. For animation control, e.g., locomotion, wave properties map naturally to stride length, step frequency, and progression, and allow smooth animation from standing, to walking, to running. Simultaneous gestures are disambiguated successfully. Interpolating out-of-phase locomotions is hard, e.g., quadruped legs between walks and runs, so we introduce a new time-interpolation scheme to reduce artifacts. These improvements to real-time motion-tracked character control are particularly important for common cyclic animations, which we validate in a user study, with versatility to apply to part and full body motions across a variety of sensors.

PDF (9 MB)
  Supplemental Material
PDF (4 MB)
PPTX (150 MB)

  Supplemental Video
MP4 (120 MB)
Extra Comparisons
MP4 (60 MB)

We thank Gabi Kussani, our professional Hohnsteiner puppeteer, our animators Gottfried Mentor and Cynthia Collins, Hung Vodinh and Joel Anderson for the horse character, Pakie Seung for the dog character, Harry Gladwin-Geoghegann for the dinosaur character, Yeongho Seol for his correspondence and his dinosaur animation, Gregorio Palmas and Hendrik Strobert for visualization help, and Michael Neff, Takaaki Shiratori, Kiran Varanasi, Simon Pilgrim, and all reviewers for their valuable discussion and feedback. This research was partially funded by the ERC Starting Grant project CapReal (335545). Kwang In Kim thanks EPSRC EP/M00533X/1. James Tompkin and Hanspeter Pfister thank NSF CGV-1110955.