This document proposes a new framework called 3DTI Amphitheater for live broadcasting in 3D tele-immersion environments. It introduces a more manageable virtual space with hierarchical stream prioritization. Performers are on a virtual stage while audiences have assigned virtual seats with manageable view directions. Stream priority is determined based on the viewing perspective and roles of performers and audiences. Evaluation shows the framework improves request admission rates and quality of service compared to previous methods. Future work includes further testing audience interactivity and dynamic role-based priority determination.

1. 3DTI AMPHITHEATER
A Manageable 3DTI Environment with Hierarchical Stream Prioritization
Shannon Chen
Klara Nahrstedt
Indranil Gupta
University of Illinois
at Urbana-
Champaign

2. 3D TELE-IMMERSION
• 3D virtual reality
• Interactive
• Free view point
• Multi-user

3. IP-BASED LIVE
BROADCASTING
• Existing IP-based broadcasting frameworks
• 2D
• Single producer at any given time
• Single view
• View decided by producer

4. 3DTI + LIVE BROADCAST
• Challenges
• High bandwidth demand
• Low latency demand
• Subscription dynamics
• Synchronization
• Content dissemination

5. 3DTI + LIVE BROADCAST
• Challenges
• High bandwidth demand
• Low latency demand
• Subscription dynamics
• Synchronization
• Content dissemination

6. 3DTI + LIVE BROADCAST
• Challenges
• High bandwidth demand
• Low latency demand
• Subscription dynamics
• Synchronization
• Content dissemination

7. 3DTI + LIVE BROADCAST
• Challenges
• High bandwidth demand
• Low latency demand
• Subscription dynamics
• Synchronization
• Content dissemination

8. 3DTI + LIVE BROADCAST
• Challenges
• High bandwidth demand
• Low latency demand
• Subscription dynamics
• Synchronization
• Content dissemination

10. 3DTI AMPHITHEATER
Virtual
Stage
3D models of
performers
Audience
Physical Space
Non-immersive
Performer
Physical Space
Immersive

11. SYSTEM MODEL
• Performer Sites
• Audience Sites
• Session management Site

12. SYSTEM MODEL
• Performer Sites
• Immersive users
• Interact on the virtual stage
• Depth camera array, head-
mounted display, acc meter
• Low latency Demand
• Audience Sites
• Session management Site

13. SYSTEM MODEL
• Performer Sites
• Audience Sites
• Non-immersive users
• Passive observer in virtual seats
• Regular display
• Relaxed latency requirement
• Session management Site
*beep*

14. SYSTEM MODEL
• Performer Sites
• Audience Sites
• Session management Site
• One manager per session
• Dissemination network
construction
• Subscription admission
• Subscription updating
• Production registration

15. USER MODEL
• Surrounding virtual seats
• Manageable view changes
• Effective content sharing
• Omni-directional coverage
• Hierarchical Stream Prioritization
• Stream differentiation:
view-based priority
• Site differentiation:
role-based priority
Source: www.stub.com/jamestown-mall-universoul-circus-tickets/

16. VIRTUAL SEATS
• Manageable view changes
• Fixed position
• Changeable direction
• Virtual opera glasses
• Effective content sharing
• Omni-directional coverage

17. VIRTUAL SEATS
• Manageable view changes
• Fixed position
• Changeable direction
• Virtual opera glasses
• Effective content sharing
• Omni-directional coverage

18. VIRTUAL SEATS
• Manageable view changes
• Fixed position
• Changeable direction
• Virtual opera glasses
• Effective content sharing
• Omni-directional coverage

19. VIRTUAL SEATS
• Manageable view changes
• Effective content sharing
• Pre-assigned seats
• Evenly spread
• Overlapping views
• Omni-directional coverage
Source: homepages.inf.ed.ac.uk/rbf/CVDICT/cvd.htm

20. VIRTUAL SEATS
• Manageable view changes
• Effective content sharing
• Omnidirectional coverage
• Every camera of a performer
is subscribed by some
audience
• ‘Hubs’ that alleviates the
dissemination burden
Source: skull-the-kid.deviantart.com/

21. HIERARCHICAL
STREAM PRIORITY
• Stream differentiation
• Site differentiation

22. HIERARCHICAL
STREAM PRIORITY
• Stream differentiation – view based priority
• Not all cameras are equally important to a viewer
• Depends on the view
• View-based priority ≡ max(CF,0)
• Site differentiation

23. • Stream differentiation
• Site differentiation – role based priority
• Not all performers are equally important to a viewer
• Depends on the roles of the performer and the viewer
• Examples for determine role-based priority
• Uniform priority
• User-defined priority
• Objective priority
HIERARCHICAL
STREAM PRIORITY
A performer is a
viewer too since
they also have
to see other
performers!

24. • Stream differentiation
• Site differentiation
• Not all performers are equally important to a viewer
• Depends on the roles of the performer and the viewer
• Examples for determine role-based priority
• Uniform priority
• Scenario: sport game
• Viewer: player1
• Performers:
{player2, oppo1, oppo2}
• Role-based priority:
{3.33, 3.33, 3.33}
HIERARCHICAL
STREAM PRIORITY
Source: www.allabouttabletennis.com

25. • Stream differentiation
• Site differentiation
• Not all performers are equally important to a viewer
• Depends on the roles of the performer and the viewer
• Determination of role-based priority
• User-defined priority
• Scenario: school play
• Viewer: Ann’s dad
• Performers:
{kid1, kid2, Ann}
• Role-based priority:
{1.00, 1.00, 8.00}
HIERARCHICAL
STREAM PRIORITY
Source: www.abbyofftherecord.com/2012/07/25/board-meeting-or-school-play/

26. • Stream differentiation
• Site differentiation
• Not all performers are equally important to a viewer
• Depends on the roles of the performer and the viewer
• Examples for determine role-based priority
• Objective priority
• Scenario: cocktail party
• Viewer: guest
• Performers:
{other guests}
• Role-based priority:
{1/distance}
HIERARCHICAL
STREAM PRIORITY
Source: www.crowneventsandconferences.com.au

27. • Construction of dissemination network
• For each stream
there is a publisher
and its
subscribers
• Together they
form a
dissemination
tree
• How does the session manager construct the forest?
• Pub/Sub model
STREAM DELIVERY
MODEL
P2P
overlay

28. • Construction of dissemination network
• All trees will be sharing
the resource of
the same
P2P overlay
• So the structure
of the forest
is important
• How does the session manager construct the forest?
• Pub/Sub model
STREAM DELIVERY
MODEL
P2P
overlay

29. STREAM DELIVERY
MODEL
• Construction of dissemination network
• How does the session manager construct the forest?
• Pub/Sub model
Publisher SubscriberBroker
Registration Subscription request
Info of the receiver Info of the sender
Establish connection
Stream
Publisher:
performers
Subscriber:
viewers
(audience +
performers)
Broker: session
manager Forest Planning

30. STREAM DELIVERY
MODEL
• Construction of dissemination network
• How does the session manager construct the forest?
• Pub/Sub model
• Registration
• Cameras
• Shooting angles
Publisher SubscriberBroker
Registration Subscription request
Info of the receiver Info of the sender
Establish connection
Stream
Forest Planning

31. STREAM DELIVERY
MODEL
• Construction of dissemination network
• How does the session manager construct the forest?
• Pub/Sub model
• Registration
• Subscription
• Site type
• Position/direction
• Priorities
• Capabilities
Publisher SubscriberBroker
Registration Subscription request
Info of the receiver Info of the sender
Establish connection
Stream
Forest Planning

32. STREAM DELIVERY
MODEL
• Construction of dissemination network
• How does the session manager construct the forest?
• Pub/Sub model
• Registration
• Subscription
• Forest planning
• Role-based priority
• View-based priority
• Bandwidth
• Latency
(…more details in the paper)
Publisher SubscriberBroker
Registration Subscription request
Info of the receiver Info of the sender
Establish connection
Stream
Forest Planning

33. EVALUATION
• Experiment 1: overall performance
• Experiment 2: service quality of performers
• Experiment 3: effect of virtual seats
• Settings
• Network: Netmap database
• 3DTI sites: TEEVE prototype
• Metrics
• Request rejection ratio
• AQoS: admission rate weighted by stream importance

34. EVALUATION
• Experiment 1: overall performance
• Compare to 4D TeleCast [2012]
• Amphitheater sustains 1,010 more requests and
provides higher AQoS
0%
2%
4%
6%
8%
10%
100
200
300
400
500
600
700
800
900
1000
Number of audiences
Reject Ratio
Amphitheater 4D TeleCast
0.9
0.92
0.94
0.96
0.98
1
Number of audiences
AQoS
Amphitheater 4D TeleCast

35. EVALUATION
• Experiment 2: service quality of performers
• Compare to Nahrstedt et al. [2011]
• x2.8 AQoS for virtual play and x1.4 AQoS for sport arena
0
0.5
1
3 5 7 9 11 13 15 17 19
AQoS
Number of participating sites
Virtual Play (User-defined)
Amphitheater
Nahrstedt et al. [8]
0
0.5
1
3 4 5 6 7 8 9 10
AQoS
Number of participating sites
Sport Arena (Uniform)
Amphitheater
Nahrstedt et al. [8]

36. EVALUATION
• Experiment 3: effect of virtual seats
• Amphitheater w/ and w/o audience
• Substantial gain in both admission ratio and AQoS
0
0.2
0.4
0.6
0.8
1
3 5 7 9 11 13 15 17 19
Requestrejectionratio
Number of performers
w/ audience
w/o audience
0
0.2
0.4
0.6
0.8
1
3 5 7 9 11 13 15 17 19
AQoS
Number of performers
w/ audience
w/o audience

37. FUTURE WORK AND
CONCLUSION
• Proposition of 3DTI Amphitheater: a new 3DTI live
broadcast framework, which introduce a more
manageable 3DTI environment
• Identification of role-based priority, which improves the
utilization by granting resource to streams that are
semantically important
• Future directions
• Stress testing: intensive view change
• Seat changing: tradeoff efficiency and user freedom
• Audience churn: seat reassignment
• Role-based priority determination: ML-based

38. THANK YOU