Loading presentation...

Present Remotely

Send the link below via email or IM


Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.


Eye tracking paper presentation

Eye Tracking for Avatar Eye Gaze Control During Object-Focused Multiparty Interaction in Immersive Collaborative Virtual Environments

Elias Giannopoulos

on 28 April 2010

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Eye tracking paper presentation

Eye Tracking for Avatar Eye Gaze Control During Object-Focused Multiparty Interaction in Immersive Collaborative Virtual Environments William Steptoe, Oyewole Oyekoya, Alessio Murgia, Robin Wolff, John Rae Estefania Guimares, David Roberts, Anthony Steed 1. Introduction perceived quality of communication Three-party scenario

Three networked Immersive Projective Technologies (IPT)

Object-focused interaction

Eye Gaze
2. Related Work 2.1 Small Group Interactions in ICVEs

IPTs provide enhanced collaboration due to intuitive head and hand tracking, as well as larger FOV.

IPTs provide enhanced task performance to desktop systems.


difficulties presented in negotiation tasks due to avatars' absence of facial expression, and understanding intentions. Capture of non-verbal behaviour is essential for remote interactions 2.2 Avatars

Avatar representations in ICVEs provide:

determination of position


visualisation of focus of attention

recognition of gestures and actions Visual and Behavioural Fidelity

Consistent realistic behaviour is beneficial to various degrees of visual realism 2.3 Eye Gaze Control

Requirement for natural communication in visual remote collaboration and conferencing systems.

Bidirectional channel monitoring initiation, maintenance and termination of messages.

Behavioural modelling - Gaze models

Simulating naturalistic eye movement for virtual characters.

Input paramters include:

fixation point and duration

saccade magnitude and velocity Left: Avatars used in Schroeder et al.'s 2001 study
(courtesy of Ralph Schroeder).
Right: Avatars used in current study. 3. Motivation and Hypothesis

Enhance remote interaction in ICVEs
more similar to face-to-face interactions Compare three methods of eye-gaze control:

Static gaze: centred eyes with no gaze control

Gaze Model: a simulation of gaze behaviour

Tracked Gaze: head-mounted eye tracking Hypothesis:

tracked gaze will result in higher quality of communication as measured by:

task performance

subjective user experience

interactional analysis
4. Technical Description 4.1 ICVE Platform: EyeCVE

EyeCVE is a tracked-gaze ICVE platform built on OpenSG

Audio communication held over Skype external to EyeCVE Binocular eye trackers mounted on IPTs modified shutter glasses

Scene camera mounted on modified shutter-glass frame (150o FOV)

Users' foveal fixation point is overlaid on the scene-camera video for real-time observation and post session analysis of gaze behaviour.

Latency from eye tracker to graphical update
on remote clients = 150ms 4.2 Eye Gaze Model

Tailored model to properties of the VE scenario

Take into account a array of objects and avatars in a user's current FOV the gaze model:

determine participant's FOV from head centric vector as defined by the IPTs head tracker (as 70o from the vector).

Randomly distribute saccades and fixations between targets (cubes and avatars' eyes) within the current FOV.

Fixation duration is dependent in timing and velocity of head movement.

5. Experimental Design Three conditions for the experiment:

tracked gaze
gaze model
static gaze

Within-subjects, repeated measure design, over the three conditions (resequenced)

Twelve males (no previous IPT experience)

Two confederates - one participant.

Participants were asked to follow a defined series of instructions issued by the two confederates.

Task: Arrange eight cubes to form a single larger cube (simple Rubik's cube)

Instruction pair:
grab instruction
position instruction 5.1 Virtual Environment

Three VEs featuring different starting configurations

VE presentation order remained constant

VEs volumes approximately equal to IPT (3x3x2.2m)

Each VE populated by 8 puzzle cubes and 5 spare ones

Three puzzle cubes where initially configured correctly 5.2 Procedure

Initial greeting of the participant by the confederates

Short training session on navigation and object manipulation

The participants were told that they would be guided by the confederates

Instructions defined as a pair of a grab and a position instruction for each of the five missing cubes
Confederates were limited to gesticulation using only their eyes and heads

Confederates were able to move freely, but only between instructions

Restricted verbal communication to:

"Pick up THIS cube" - (grab)

"Place the cube HERE" - (position)

5.3 Recorded Data

Video and Audio: recorded from eye tracker scene cameras worn by participant and both confederates

Replay Logs

Eye tracker Logs

Gaze Model Logs

After sessions' questionnaire: subjective experience, co-presence, self-performance

Informal post experimental questionnaire

6. Results Condition Static Model Tracked

Grab Errors 1.67 (1.44) 3.42 (1.78) 1.67 (1.37)
Grab Time 48.0 (22.3) 68.1 (34.2) 47.6 (20.6)
Position Errors 0.75 (0.87) 0.50 (0.90) 0.47 (0.67)
Position Time 24.2 (18.5) 26.6 (14.3) 20.9 (9.7)
Total Time 270.1 (67.2) 301.9 (133.4) 310.3 (132.8) Task Performance

User Experience

Interactional Analysis 6.1 Task Performance

Five Pairs of Grab and Position Instructions

Grab Instructions Errors and Timing

Position Instructions Errors and Timing

Puzzle Completion Time 6.1.1 Grab Instruction Errors and Timings

Two-way analysis of variance (ANOVA) for grab instructions errors and timings

Three gaze conditions and five grab instructions as factors

a. Grab instruction errors as well as timings: significant difference between conditions and also between instructions

b. Differences between gaze model and the other two conditions

c. No interaction between eye gaze condition and grab instruction Tracked and static gaze resulted in significantly fewer errors and faster operation time than the gaze model while performing grab instructions. 6.1.2 Position Instruction Errors and Timings

Two-way analysis of variance (ANOVA) for position instructions errors and timings

Three gaze conditions and five position instructions as factors

a. Position instruction errors and timings: no significant difference between conditions but significant between instructions

b. Differences between gaze model and the other two conditions

c. No interaction between eye gaze condition and position instruction No significant differences between conditions regarding errors and time while performing position instructions 6.1.3 Puzzle Completion Time

One-way analysis of variance (ANOVA) for overall puzzle completion time

No significant difference between conditions.

Subsequent analysis:
Completion time - Experience

One-way ANOVA on each session's puzzle completion time

Using session order as factors

Significant difference between the orders was found. 6.2 Subjective User Experience

Two-way ANOVA taking eye gaze conditions and the fourteen questions.

No significant difference between conditions.
Although not statistically significant, users rated the tracked-gaze session higher than the static gaze ones and in particular the gaze model.

This was supported by the post-experimental interview.

Under all conditions judgement of experience and self-performance ratings were consistently high. 6.3 Interactional Analysis

Responses to Grab Instruction

Movement as a Resource

Eye Gaze as a Resource

Glances per Grab Instruction 6.3.1 Responses to Grab Instruction

Confederate typically turns and looks at the designated cube when the deictic term 'this' is uttered.

Participant follows confederate's gaze, and glances between cubes and confederate.

Once he identifies the cube looks at the cube and touches it while requesting confirmation and gazing towards the avatar.

Repair strategies:

Participant looks from one confederate to another to establish which one is talking (no mouth movement)

Confederate moves and gazes towards the participant to establish the connection.

Participant shifts gaze towards potential cubes - glancing between avatar and cubes.

Finally participant moves closer to speaking avatar to examine his gaze.

6.3.2 Movement as a Resource

Confederates proximity to cubes - useful to participants

Confederates position themselves at appropriate distance and clear angle to intended cube.

In many cases participants take up an opposite position to the avatars' behind the cubes.

Participants made several uses of their freedom of movement. 6.3.3 Eye gaze as a Resource

Participants shares their gaze between both confederates between silences, conversations and instructions

The gaze model proved to be quite confusing, causing difficulties, which the participants expressed as questions or complaints.

6.3.4 Glances Per Grab Instruction

A strategy adopted by all participants and for 98.9% of grab instructions, was to alternate the gaze between instructing avatar and potential cubes.

Fewer glances before responding to grab instruction:
reduced time required per instruction
ability of eye gaze control method to indicate objects

Two-measure ANOVA performed for glances per grab instruction, against the three eye-gaze conditions and the five grab instructions as factors.

Significant difference between conditions and between instructions. Condition Static Model Tracked
# of Glances 2.42 (1.5) 3.58 (2.74) 2.25 (1.57) 7. Discussion 7.1 Naive Users and Task Complexity

Inexperienced IPT users

Requirements: communication, navigation, judgement calls, object manipulation, questionnaires

Sessions lastes up to 2 hours.

Many factors influencing participant performance.

However poorer performance of gaze model indicates significance of eye gaze.

Head orientation and body positioning provides sufficient nonverbal communication.

7.2 Avatar Representation

Single avatar model

Photographic textures

Low fidelity in terms of geometry and realism

Lack of eyelid movement and blinking

Lack of mouth movement (lip-synchronisation)

7.3 Gaze Model Performance

Percentage equality of eye-target (ET), head-target (HT) and model-target (MT) in all sessions. Small points denote individual participants. Square is mean. Range is standard deviation. 8. Conclusions

Tracked and static gaze avatars were seen to support significantly higher quality of communication than avatars exhibiting gaze model behaviour.

No statistical significant performance between static and tracked gaze, although tracked gaze scored higher.

Future Work

Development of effective and reusable gaze models

Demonstrate tracked gaze superiority through redirected experimental and avatar design.

Eye Gaze

Critical aspect of avatar design

Levels of behavioural and representational fidelity of avatar eyes ICVEs
Immersive Collaborative Virtual Environments Connect remote users within:
informational context

Support high quality interaction Participant representation as avatars

Head orientation (head tracker)
Pointing gestures (hand tracker)

Shared computer generated virtual environment (VE) objective and subjective performance measures during object identification tasks Thank you!
Full transcript