Motion Modeling for OnLine Locomotion Synthesis

1
Motion Modeling for On-LineLocomotion Synthesis

• ACM SIGGRAPH / Eurographics Symposium on Computer
  Animation 2005

2
Motion Modeling for On-LineLocomotion Synthesis

• Taesoo Kwon (???)
• Sung Yong Shin (???)
• Korean Advanced Institute ofScience and
  Technology

3
Introduction

• Motion Synthesis
• On-line, real-time
• Blending
• Efficiency and controllability
• Kovar et al., Automated extraction and
  parameterization of motions in large data sets,
  2003
• Rearrangement
• Naturalness
• Kovar et al., Motion Graphs, 2002

4
Introduction

• A hybrid approach
• Blending Rearrangement
• Motion transition graph
• NodeA group of basic motions of identical
  structure for blending
• EdgeTransition from a blended motion to a
  blended motion

5
Overview

• Motion Analysis
• Example motion preprocessing
• A labeling scheme Segmentation and
  Classification
• A hierarchical motion transition graph Coarse
  level and Fine level

6
Overview

• Motion Synthesis
• Given specifications in an on-line manner, the
  motion transition graph is traversed to
  synthesize the motions.
• Parameterization, weight computation,time-warping
  and posture blending
• Park et al., On-line motion blending for
  real-time locomotion generation, 2004

7
Outline

• Motion Analysis
• Segmentation
• Classification
• Parameterization and Keytime Extraction
• Motion Transition Graph Construction
• Motion Synthesis
• Specification
• Blending
• Results, Discussion and Conclusion

8
Motion Analysis

• Segmentation
• Classification
• Parametrizaiton and Keytime Extraction
• Motion Transition Graph Construction

9
Motion Segmentation

• Biomechanical observations

10
Motion Segmentation

• Identify all standing motion segment by detecting
  every frame with a stand pose.
• Cut the remaining portion into motion segments
  with peaks.
• S s0, s1, , sk such thatM s0 s1
  sk

11
Motion Analysis

• Segmentation
• Classification
• Parametrizaiton and Keytime Extraction
• Motion Transition Graph Construction

12
Motion Classification

• Five footstep patterns
• Stand phase (S)
• Left foot support phase (L)
• Right foot support phase (R)
• Double support phase (D)
• Flight phase (F)

13
Motion Classification

• Motion labelingEncode each motion segment as a
  string of symbols in ? L, R, D, F, S .
• S fixed COM position
• D both feet contact the ground
• F feet are above the ground
• L and R only one foot contacts the ground

14
Motion Classification
15
Refinement

• High-to-low speed transition
• A small COM peak may occur
• Concatenate by ignoring the peak
• Low-to-high speed transition
• Missing COM peak
• Split into two parts
• The remaining invalid stringsare discard.

16
Motion Analysis

• Segmentation
• Classification
• Parametrizaiton and Keytime Extraction
• Motion Transition Graph Construction

17
Prametrization

• The objective for motion parametrizaiton is
  on-line locomotion control.
• For a motion segment m,define a parameter
  vectorp(m) ( t(m), ft(m), v(m),?(m), a(m) )
• t(m) motion typeft(m) contact status
• Directly obtained from the table

18
Prametrization

• COM trajectory the projection
  of onto the ground
• ,
• Let be a maximal subsequence of
  frames in m such that for all

19
Keytime Extraction

• Every keytime of a motion segment is a moment at
  which a motion feature occurs.
• Keytimes are used to time-warp the motions to be
  blended for feature alignment.
• The keytimes in locomotion are the moments of
  heel-strikes and toe-offs.
• The start and end frames of everyL or R phase
• Every COM peak

20
Motion Analysis

• Segmentation
• Classification
• Parametrizaiton and Keytime Extraction
• Motion Transition Graph Construction

21
Motion Transition Graph Construction

• Coarse level Fine level

22
Motion Synthesis

• Specification
• Blending

23
Motion Specification

• A motion specification prescribes only the
  locomotive motions at coarse-level nodes.
• The motion specifications for basic movements are
  derived automatically from the motion context.

24
Motion Context

• Three cases
• Standing motionft(m),?(m) and a(m) are trivially
  determined, since there are no footsteps, no
  turning and no acceleration.
• Cyclic motion, such as walking or running
• Transition motion

25
Cyclic Motion

• Let m be the basic movement synthesized at the
  current step.
• If the previous basic movement is not available,
  ft(m) is chosen at random
• ft(m) is left if ft(m) is right, and vice versa.

26
Cyclic Motion

• Empirically, it turns out that, is
  correlated with and .
• By regression analysis,

27
Transition Motion

• The motion type is obtained from the table.
• If m is not a standing motion, then the other
  parameters are simply copied from those of m.
• Otherwise, those parameters are generated in a
  similar manner asfor a cyclic motion by ignoring
  m.

28
Motion Synthesis

• Specification
• Blending

29
Motion Blending

• Park S. I., Shin H. J., Shin S. Y. On-line
  motion blending for real-time locomotion
  generation, Computer Animation and Virtual
  Worlds, 2004
• The framework is composed of four parts
• Parametrization
• Weight computation
• Time-warping
• Posture blending

30
Weight Computation

• Inverse distance weighted interpolation
• k-nearest neighbors interpolation
• Let the k nearest neighbors be p1,, pk , in
  order of increasing distance, and let wi be the
  blended weights with pi, then the blend weight w
  for the new set of parameters p can be found as

31
Time-Warping

• Incremental time-warping
• Blend the change rates of the actual times with
  respect to the generic time rather than the
  actual times.

32
Time-Warping
33
Outline

• Motion Analysis
• Segmentation
• Classification
• Parameterization and Keytime Extraction
• Motion Transition Graph Construction
• Motion Synthesis
• Specification
• Blending
• Results, Discussion and Conclusion

34
Results
35
Results

• Path-following
• Motion type color of the curve
• Speed tangent vector at the current point
• Turning angle angle between the tangent vectors
  at the previous and current points
• Joystick
• Motion type button
• Speed and turning angle stick position
• Mouse
• Motion type keyboard
• Speed and turning angle cursor position

36
Discussion

• Their labeling scheme is extremely robust to
  avoid any human adjustment except
• The unlabeled example motion should be captured
  from a real performer, but not created manually
  by an animator.

37
Conclusion

• An example-based, on-linelocomotion synthesis
  method
• The labeling scheme
• Biomechanical observation
• String processing
• Future work
• Dancing, boxing and martial arts

38
End

• Presented by Ying-Shou Lan