Parameterized Motion Interpolation

1
Parameterized Motion Interpolation
Extrapolation Using Partial Blending

• Student Yan-Ju Chen
• Advisor Prof. I-Chen Lin

2
Outline

• Introduction
• Related Work
• Overview
• Preprocess Phase
• Runtime Tuning Phase
• Experiments Results
• Conclusion Future Work

3
Introduction

• Motivation
• Acquisition of many data from motion capture
  costs a lot
• Usually captured data are not well-designed and
  somewhat imperfect
• Idea
• Human body can be segmented into several parts
  and then spliced back!

4
Introduction

• Goal
• Make more use of the limited fighting action data
• Extend and fix their possible attack range
• More directions and positions to attack
• Interactive and real-time control
• Gaming

5
Related Work

• Motion Graphs SIGGRAPH 02
• Find out proper transition points

6
Related Work

• Automated Extraction and Parameterization of
  Motions in Large Data Sets SIGGRAPH 04

7
Related Work

• Splicing Upper-Body Actions with Locomotion
  EUROGRAPHICS 06

8
Overview
9
Preprocess Phase Motion Parameterizing
10
Calculate Distance Between Frames

• Two point clouds indicate joint positions

11
Time Alignment

• Grid form
• Find alignment path
• Dynamic programming
• Continuous
• Monotonically increasing

12
Parameter Extraction

• Denser sampling
• Apply various weight combination to blend more
  motions
• Slerp
• Record their attack position as parameters
• Normalize to a parameter plane

13
Parameter Extraction
14
Determine Attack Range

• Find the largest and smallest value in x and y
  direction
• Extend possible attack region
• A field whose boundaries are the values
  discovered above

15
Runtime Phase Motion Tuning
16
Segment Human Body Into Parts
17
Query Parameters

• Search for two parameters in the original
  parameter space
• One closest in horizontal direction
• One closest in vertical direction

A
B
18
Angular Adjustment

• Principal
• Use the major active body part style of A
• Take the direction of B as reference
• General
• We wish to give different angular adjustments to
  different body segments so as to attack the
  desired target
• Frame by frame angular momentum implicit

19
Angular Adjustment
A
B's orientation
A's orientation
Adjust
GOAL
B
20
Angular Adjustment

• Two cases
• Differs in major active body part
• Punching motion right hand
• Kicking motion right leg

21
Angular Adjustment

• Punching motion
• Hierarchical
• Hips
• Chest
• RightCollar

22
Angular Adjustment

• Kicking motion
• Hips and RightHip adjustment
• Extra lift up or lay down on right leg

23
Rearrange Motion Length

• Smoothen motion
• Due to many-to-one correspondence in time
  alignment step
• According to the largest weight of the example
  motion

24
Experiments Results

• Environment
• Intel Pentium 3.20 GHz CPU
• 1.5 GB RAM
• NVIDIA GeForce 6600
• Example data
• 3 kinds of punching motion
• 2 kinds of kicking motion

25
Experiments Results

• Video

26
Conclusion Future Work

• Conclusion
• An efficient example-based method to control
  motion capture data of fighting actions
• Create reliable motions that attack the desired
  target
• Extend possible attack range of the character gt
  enrich the usage of the original imperfect
  captured data

27
Conclusion Future Work

• Conclusion
• Simple interactive control can be used in
  real-time applications, ex games
• Can deal with not only fighting actions but also
  similar motion which can be parameterized

28
Conclusion Future Work

• Future work
• For more accuracy
• 2D to 3D parameterization
• For more possibility
• Attack path
• Further physics-based adjustment

29
Thanks