Group Motion Editing

SIGGRAPH 2008

Presented by Ting-sheng Lin

Taesoo Kwon1 1Seoul National

University

Kang Hoon Lee2

2Kwangoon University

Jehee Lee3

3Seoul National

University

Shigeo Takahashi4

4University of Tokyo

Introduction

Background

Graph Construction

Editing Group Motions

Stitching Group Motions

Postprocess

Experimental Results

Discussion

4

Crowd Scenes

Feature animation films

Video game

State of the art

Simulating each individual that chooses its action

Rules, decision models, force fields

Careful parameter tuning

Lack precise control of individuals

Laborious trial and error

5

Interactive editing scheme

Complements such simulation-based techniques

Animators have direct control over animated crowd behaviors

Selectively edit and combine some portions of the simulation

results to achieve globally satisfactory results

Manipulating the motion of multiple characters

Repositioning only a few characters

Preserve each individual trajectory and group formation of

individuals

6

Novel graph structure

Each vertex: location of an individual at a sampled frame

Connecting edge: individual moving trajectories and

neighborhood formation

Mesh editing method [Igarashi et al. 2005]

Deform a group motion

Stitch two independent group motions

Avoiding collisions

7

Synthesizing realistic group

Agent models

[Reynolds 1987; Musse and Thalmann 1997; Pelechano et al. 2005,

Shao and Terzopoulos 2005]

Directly model

[Chenney 2004, Hughes 2003; Treuille et al. 2006]

Data-driven method of constructing group behavior models

[Lai et al. 2005; Lee et al. 2007; Lerner et al. 2007; Courth and Corpetti

2007; Paris et al. 2007]

9

Motion editing

Motion clips to be manipulated with constraints

[Gleicher 1997; Lee and Shin 1999]

Concatenated [Rose et al. 1996]

Interpolated [Rose et al. 1998; Mukai and Kuriyama 2005]

Rearranged [Lee et al. 2002; Kovar et al. 2002; Arikan et al. 2003]

Our goal is to provide users with a similar level of flexibility in

edition group motions

10

Our approach

Data-preserving shape editing [Igarashi et al. 2005]

Maintaining the local arrangement of vertices allows user to

intuitively manipulate 2D and 3D shapes

As-Rigid-As-Possible Shape Manipulation [Igarashi et al. SIGGARPH

05]

11

Large crowd animation

Tractable motion clips

Each group motion clip consists of a set two-dimensional moving

trajectories of individual characters

Preserving spatial relations among individuals

13

.

Formation edge

Represent the neighborhood relationships between individuals

Challenge: correctly identify the formational relationships

Neighborhoods can vary according to time

Don’t keep pace with other characters and moves at variable speeds

Delaunay triangulation of vertices at each plane

Motion edge

Connect with the vertex of the corresponding character at the

previous plane and next plane14

NjjiV 1, }{ ],...,2,1[ Ti

16

It is invariant on the uniform scaling of local features

Unnaturally enlarged, shrunken

17G

G’

Two-step optimization scheme [Igarashi et al. 2005]

Step one: scale-free construction

Step two: scale adjustment

Graph undergoes a large deformation

Highly-distorted

Near-degenerate triangles

Group motion editing

Motion artifacts: sudden velocity changes

Spatial group formation: scale-adjustment

Temporal distortion: time-warping18

We consider three types of triangular features

Spatial, temporal, spatiotemporal features

19

Vt

Vt+1

Vt-1

wt

st

ut

Vt

Vt+1

Vt-1

wt

st

ut

Vt

Vt+1

Vt-1

wt

st

ut

Spatial feature

temporal feature

spatiotemporal feature

2

''

')','()()'(

sCc

S vwufccwGE

20

2

''

')','()'(

TCc

T vwufcGE

2

''

')','()'(

STCc

ST vwufcGE

)/(1)(22

vwvucw

Only the spatial features are scaled

Construct a spare linear system that can solve E1 and E2 Variable elimination [Igarashi et al. 2005]

Lagrange multiplier scheme

21

G and G’ having the same number of individuals N

Stitching two graphs requires three steps

First step

23

G G’

Bipartite graph matching algorithm [Belongie et al. 2002]

G

G’

Second step

Final step: smoothly morphing group formations

Linear blending don’t generate desired results

We blend triangular (spatial, temporal, spatiotemporal)

24

Aligned two motion clips by translating and rotating them to best match the boundary [Kovar et al. 2002]

G G’

G’’

Deformation or stitching of a group motion

Collision, insufficient clearance

Collision avoidance algorithm: approximate and iterative method

Certain threshold: two trajectories

Pulls the trajectories away by 10% of the threshold and repeats

this process until all collision are resolved

Avoid collision with obstacles We pull the deepest penetrating point

toward the nearest point on the boundary

of the obstacle

26

This speed change is often irregular

Irregular speed change is usually undesirable in motion editing

Time warping

27

29

1. Battle field scene

2. Downtown scene

Conclusions

Animators to manipulate existing group motion data interactively

Detail-preserving approach

Advantage: direct, precise control

Limitations

Cannot large deformation, but lead to unnatural speedup/slowdown

Cannot handle a large crowd consisting of thousands character

Future work

Deal with a wider variety of group behaviors (e.g: chatting, Olympic

…)31

32