Motion Graphs

1
Motion Graphs

• Lucas Kovar
• Michael Gleicher
• Frederic Pighin
• University of Wisconsin-Madison
• University of Southern California

SIGGRAPH 2002
5. 9. 2006 - Presented by Nahyup Kang -
2
Outline

• Introduction
• Related Works
• Background
• Motion Graph Construction
• Extracting Motion
• Path Synthesis
• Conclusion

3
Introduction

• Realistic Human Motion is very important
• Video game
• Movie
• Motion Capture
• Good solution
• Hard to reuse
• Modify
• Edit

4
Introduction

• Goal
• Retain the realism of motion capture
• Give a user the ability to control and direct a
  character
• Example
• With general walking motion sequence
• Generate special path of walking motion

5
Related Work

• Editing individual clips of motion
• Retargeting Gleicher 1998 Lee and Shin 1999
• Multi-target blends
• Linear Interpolation Wiley and Hahn 1997
• Radial Basis Function Rose et al. 1998
• Move Trees Mizuguchi et al. 2001
• Game industry
• Manual Creation
• Motion Transition
• Perlin 1995 Rose et al. 1996 Lee 2000

6
Background

• Motion Capture Data Structure

Root Joint End Effector
7
Motion Capture Data Sample
HIERARCHY ROOT Hips OFFSET 0.00 0.00 0.00
CHANNELS 6 Xposition Yposition Zposition
Zrotation Xrotation Yrotation JOINT LeftHip
OFFSET 3.29 0.00 0.00 CHANNELS 3
Zrotation Xrotation Yrotation JOINT
LeftKnee . MOTION Frames 30
Frame Time 0.033333 10.87 36.65 13.54 8.70
1.67 91.18 10.89 9.41 -1.80 -2.30 7.50 12.08 0.00
-10.56 0.00 -13.95 10.19 2.52 1.45 7.58 -10.90
0.00 -12.38 0.00 -2.30 7.22 -0.10 -0.13 0.00 5.48
48.69 -5.16 13.44 -10.40 -22.07 -14.50 -0.21
-5.66 -0.02 7.72 0.00 -7.17 -52.62 2.63 -1.95
-1.90 -20.04 -18.84 0.16 -4.85 0.01 0.00 35.36
0.00 1.77 -36.47 -2.39
8
Background

• General Representation
• F Number of Frames
• J Number of Joints
• t Frame Index

9
Motion Graph Construction

• Motion Graph
• Directed Graph
• Edge
• Correspond to clip of motion
• Node
• Choice point connecting these clips

10
Motion Graph Construction
Where is the proper point to make an edge?
11
Detecting Candidate Transitions

• Find similar frames
• How about vector norm of frame?
• Priority of the joints are ignored
• No meaningful way to assign fixed weights
• Identifying compatible coordinate systems are
  necessary
• Joint velocities, accelerations, higher-order
  derivatives are necessary.

12
Pose Distance Metric
13
Selecting Transition Points

• Local minimum in the distance function
• Transition better than its neighbors
• Local minimum with small error values
• Candidate Transition Points

14
Creating Transitions

• Blending frames to with frames
  to
• Aligning
• On frame p linearly interpolate
• Blending Weight?

15
Creating Transitions

• Linear blending
• Easy and simple and fast
• The constraints in the motion can be violated
• Foot sliding (footskate)
• We have to correct this violation
• Treat constraints as binary signal
• Blending this signal
• Enforce this blended constraints as a
  post-processing step

16
Pruning The Graph

• Can Motion Graphs synthesize motion indefinitely?
• NO! There may be deadend!

• Compute Strongly Connected Components of graph
• Remove any edge that does not attach two node in
  the largest SCC.

17
Extracting Motion

• How to play the motion?
• Exactly How to traverse the Graph?
• Converting Graph Walks to Motion

18
Converting Graph Walks To Motion

• Every edge on the motion graph is a piece of
  motion.
• Aligning correct location and orientation of
  pieces
• Footskate clean-up Kovar et al. 2002

19
Searching For Motion

• Problem of finding motion that satisfies
  user-specified requirements
• Graph Search Problem
• Metric
• Time elapsed
• Distance Traveled
• g(w,e) user define nonnegative error evaluator
• Find the path that minimize f
• Branch and Bound

20
Deciding What To Ask For

• g should
• give some sort of guidance throughout the entire
  motion
• be no more restrictive than necessary
• be balanced.

21
Path Synthesis

• Implementing Path Synthesis
• Cost Function Design
• Type Change

22
Path Synthesis

• Application Of Path Synthesis
• Interactive Control
• Go left!
• Make left path immediately.
• Pre-compute a sequence of graph walks that fits
  straight line path
• High-Level Keyframing
• Path with appropriate action type
• Back-end for animating
• The path and action type can be specified by a
  high-level process
• Motion graph fill in the detail
• Crowds
• Can be applied to several characters in parallel.

23
Path Synthesis

• Results
• Enjoy the movie clip

24
Discussion Conclusion

• Framework for generating realistic, controllable
  motion through a database of motion capture.
• Graph is automatically constructed among
  different piece of motion
• Finding the candidate transitions is bottleneck.
• Transition Threshold is manually selected.
• Motion parameterization is not considered.

25
End

• Thank you.
• Question and Comment.