GAZE Issues

1
GAZE Issues

• CS 6724
• Embodied Interaction
• Ajit Kulkarni
• 31st October 2006

2
Papers

• Rainer Stiefelhagen, and Jie Zhu, "Head
  Orientation and Gaze Direction in Meetings,
• Garau, M., et al., "The Impact of Avatar Realism
  and Eye Gaze Control on Perceived Quality of
  Communication in a Shared Immersive Virtual
  Environment,"
• Vinayagamoorthy, V., M. Garau, A. Steed, and M.
  Slater, "An Eye Gaze Model for Dyadic Interaction
  in an Immersive Virtual Environment Practice and
  Experience,".

3
Paper 1 Head Orientation and Gaze Direction

• Meetings Who is looking at whom???
• Gaze Direction where eyes are pointing in space
• Gaze Head orientation Eye orientation
• Head orientation -gt estimation is non intrusive
• Eye orientation -gt estimation is intrusive
• Eye orientation estimation requires lots of
  equipment like HMDs or restrictions on behavior

4
Goals of Paper

• Finding contribution of head orientation to
  overall gaze
• Predicting users focus of attention using head
  orientation only

5
Data Collection

• Round table meeting of 4
• Head mounted system used by subject
• Head orientation and position -gt magnetic pose
  and position tracking system
• Eye orientation and position -gt head mounted
  camera
• Gets eye blinks also-gt no eye gaze data at this
  point

6
Head Orientation Contribution To Gaze

• 87 of the time head orientation and eye
  direction in same direction
• Subjects use head orientation to change gaze
  direction
• 20 of the time eye blinks -gt no eye tracking
  -gtonly head orientation

7
Target Estimation Using Head Orientation Only

• Labeling of frames when a person was looking at
  some one else
• Find focus of attention label
• Gaussian modeling approach
• Estimate focus of attention target using head
  orientation data and compare with data from
  labels
• 82-93 of the time correct estimation of target
• Very good , we have only this during eye blinks

8
Conclusions

• Head orientation contributes extensively (avg68
  in overall gaze)
• Attention estimation using head orientation only
  -gt accuracy 88.7

9
Paper 2Impact of Avatar Realism and Eye Gaze
Control in IVEs

• Goal
• Investigate impact of visual and behavioral
  realism in avatars in quality of communication in
  immersive virtual environments
• How peoples communication experience changes to
  varying levels of visual and behavioral realism

10
Types of Gaze

• Random Gaze
• Avatars head and eye animation is unrelated to
  conversational flow
• Inferred Gaze
• Head animations tracked and eye animations
  inferred from audio flow
• Static Gaze
• Fixed look at some one for certain time

11
Hypotheses

• Behavioral Realism independent of visual realism
  on quality of communication
• Behavioral Realism is of higher importance
• Inferred gaze will out perform random gaze for
  all types of avatars

12
Experiment Setup

• Pair of same gender for 1 test
• First is in CAVE second uses HMD
• Two training rooms and a negotiation room
• Role playing negotiation task
• Low and high realism avatars used

13
Quality of Communication Indicators

• Response variables
• Face to Face
• Involvement
• Co-presence
• Partner Evaluation

14
Analysis and Discussion

• Logical regression analysis
• Strong interaction effect between type of avatar
  and type of gaze
• Inferred gaze outperforms random gaze
• Comparison of gaze models in an immersive setting
• Avatar appearance
• High realism avatar doesnt outperform low
  realism avatar

15
Conclusions

• Independent of head tracking inferred eye
  animations have a significant positive effect on
  immersive interaction
• There must be a certain degree of visual realism

16
Paper3 Eye Gaze Model for interaction in IVE

• Motivation
• Avatars used in social interaction in virtual
  environments
• Effective representation of non verbal behavior
• Virtual humans must be made more realistic
• Create behavioral model using inferences from
  real human behavior
• Model takes estimation of humans intention and
  the avatar emulates it

17
Central Idea

• Behavioral realism is important for improving
  realism and believability
• Focus on eye gaze

18
Behavioral Module Goals

• Building a adaptive framework to create a virtual
  identity using high level variables
• Study how changed in variables affect behavior of
  virtual humans
• Explore optimum levels of visual and behavioral
  realism to make the character believable

19
Behavioral Module Design

• Avatars represent participants in shared IVE
• Avatar behavior changes depending on
  participants real time speech and tracking data
• Components
• Eye Gaze Module
• Inferred gaze module
• Random gaze module
• Other body animations
• Head control
• Arm and leg movements

20
The Behavioral Module
21
Avatar States

• Talking
• Listening
• Eyes looking ahead (primary position)
• Eyes looking away

22
Eye Behavior Terms

• Primary position
• At gaze
• Away gaze
• Saccade
• Inter saccade interval
• Saccade duration
• Saccade magnitude
• Saccade direction
• Saccade velocity

23
Inferred Eye Gaze Model Assumptions

• Mean saccade magnitude
• person listening lt person speaking
• Inter Saccade interval
• person speaking lt person listening
• People look at partner more when listening than
  when speaking

24
Inferred Eye Gaze Model
25
Random Eye Gaze Model

• Same as inferred eye gaze model but
• No saccade velocities involved
• Saccade magnitude obtained by random number
  generator

26
Other Body Animations

• Head Control
• Track users head tracker in X-Z plane
• Boundary circle
• gt radius full body translation
• lt radius only head rotation
• Arm and Legs
• Track right hand and legs
• Use 3D projections on XZ plane
• Consider projection of elbow , wrist and shoulder

27
Paper Conclusions

• Effectiveness increases when realistic eye gaze
  used with realistic avatar
• Consistency needed between visual realism and
  behavioral realism

28

• Thank You