Creating Adaptive Views for Group Video Teleconferencing

1
Creating Adaptive Views for Group Video
Teleconferencing An Image-Based Approach 

• Ruigang Yang Celso Kurashima Andrew Nashel
  Herman Towles Anselmo Lastra Henry Fuchs

University of North Carolina at Chapel Hill
2
Current Teleconferencing
Transport
Display
Capture
?
3
The Office of the Future
4
Group Teleconferencing
Cameras

• Multiple persons (3-4) at each site
• Life-size, monoscopic display
• High-resolution seamless imagery
• Active view control

5
Active View Control
Provide the best approximating view
6
Active View Control

• A view synthesis problem
• Extract 3D geometry from a few cameras
• Less expensive
• Hard to get good results
• Image-based method capture many images
• Looks really good on every scene
• Need many images

7
Our Image-based Approach

• Observation Eye level remains relatively the
  same during a conference session
• A compact Light Field representation
• Parameterized by a 3D function (s, u, v)

8
Linear Light Field
9
LLF Rendering

• Projective Texture mapping and blending
• Tessellate the focal plane
• Project input images onto the focal plane
• View-dependent blending

New view
Focal Plane
Base image
10
Blending Function
Focal Plane
11
Samples Images
Perspective Projection
Orthogonal Projection (extreme case)
12
Sampling Analysis

• Configuration parameters
• Focal plane depth D
• Cameras FOV
• Cameras horizontal resolution W
• Inter-camera distance d
• Error term pixel drift (e)

Given the configuration parameters, and a desired
error tolerance e , what is the maximum depth
deviation ?D from the optimal depth D.
13
Sampling Analysis Result
14
More results

• Distributed System
• 11 cameras
• 6 capture PC ( 640x480)
• ROI encoded
• JPEG compression
• One rendering PC
• Roughly 1000 x 480 output
• 4-7 frames per second

15
Conclusions

• We presented a novel system designed specifically
  for group video-teleconferencing.
• Best approximate view for the group
• Photo-realistic results at interactive rate
• Flexible and scaleable

16
Acknowledgements

• Funding support from
• The Department of Energy's ASCI VIEWS Program
• Sandia National Laboratories USA
• Collaborators from Sandia
• Phil Heermann
• Christine Yang
• Corbin Stewart

17
The End
Thank You