University of Paris 8

1
University of Paris 8

• Animation improvements and face creation tool for
  ECAs
• Nicolas Ech Chafai
• Benjamin Dariouch
• Maurizio Mancini
• Catherine Pelachaud

2
Overview

• aiming at improving agent's facial animation
  quality
• we are studying some motion captured data
• we apply results to our ECA
• to allow the creation of individualized ECAs
• we developed one tool for MPEG4 face creation
• we propose some refinements to MPEG4 specification

3
MOCAP data analysis

• Three main goals
• displacement of FAPs during emotion presentation
• synchronization between different FAPs
• FAP values during transition between consecutive
  emotions

4
Collected MOCAP data

• 2 actors
• 33 markers, 21 of them are the MPEG4 FAPs
• 78 sequences
• basic movements
• rising eyebrows
• smiling
• basic emotions
• anger
• happiness
• surprise

5
MOCAP problems

• (we discovered that) obtaining usable data is not
  straightforward
• right size and shape markers have to be used
• cameras have to be placed properly
• data has to be translated to the needed reference
  system
• data has to be filtered from noise

6
Data example

• smile

7
Video examples

• frown clip
• file clips/coline 36 eyebrows
• fear clip
• file clips/coline 56 fear

8
Facial animation model

• FAPs displacement during basic emotions
• our model was simply based on onset-apex-offset

9
Data observed model

• on real data we observed other general behaviors

10
Results

• we started to introduce the ADSR model
• given a sequence of (phase,intensity,duration)
  where phase is from Attack, Decay, Sustain
  Release, the FAP curve is built using keyframe
  Hermite interpolation

11
ADSR vs real data
12
ADSR example

• clip

13
MPEG4 face tool
14
MPEG4 face tool

• imports models from Poser
• allows the selection of the areas influenced by
  FDPs

15
Tool's features

• automatic selection and symmetrization
• automatic association region name available
  FDPs

16
Example

• exports into a data file (containing geometry
  regions) readable from the Greta's player

17
Example

• flat.avi (note female speech)

18
Added new FAPUs

• after adding new faces some refinements for the
  MPEG4 player will be needed

19
Example

• clip without new FAPUs
• clip with new FAPUs

20
Conclusions

• more data has to be captured in proper way
• focus more on interaction between different FAPs
  and transition between sequential expressions
• ADSR has to be fully implemented
• for documentation, papershttp//www.iut.univ-pari
  s8.fr/greta
• for Greta applications available to the other
  HUMAINE members please contact usm.mancini_at_iut.un
  iv-paris8.fr