1 / 58

David Nguyen Professor John Canny

Multi V iew. A Spatially Faithful Video-Conferencing System. David Nguyen Professor John Canny. Talk Outline. Motivation Prior Work Our Approach Experiment Discussion Conclusions.

vui
Download Presentation

David Nguyen Professor John Canny

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. MultiView A Spatially Faithful Video-Conferencing System David Nguyen Professor John Canny

  2. Talk Outline Motivation Prior Work Our Approach Experiment Discussion Conclusions

  3. MultiView is the first practical video conferencing system that preserves many non-verbal cues in group-to-groupmeetings by improving spatial faithfulness

  4. Spatial Faithfulness describes how attention cues (e.g., gaze and gesture) are preserved or distorted across a virtual space.

  5. Motivation You’re Fired!

  6. Spatial DistortionsCollapsed Viewer Effect With only one camera, remote participants take on a shared and incorrect perspective warping nonverbal cues

  7. 0 10 20 35 50 Mona Lisa Effect

  8. Spatial Faithfulness • Different levels of Spatial Faithfulness • Mutual – Correctly perceive that attention is or is not being directed at you when it actually is or is not • Partial – Correctly perceive the general direction of attention when it is away from you • Full – Correctly perceive the specific object of attention increasing Dave Donald

  9. Talk Outline Motivation Prior Work Our Approach Experiment Discussion Conclusions

  10. HYDRA(Sellen et al., 1992) Group Sites Mutual Sp.Fa. Partial Sp.Fa. Full Sp.Fa.

  11. GAZE-2 (Vertegaal et al., 2003) Group Sites Mutual Sp.Fa. Partial Sp.Fa. Full Sp.Fa.

  12. MAJIC(Okada et al., 1994) Group Sites Mutual Sp.Fa. Partial Sp.Fa. Full Sp.Fa.

  13. Talk Outline Motivation Prior Work Our Approach Experiment Discussion Conclusions

  14. Unique and Correct Views “Virtually Here” Multiple Viewpoint Display Preserving a geometric relationships between virtualand actualpositions provides full spatial faithfulness

  15. Gaze Parallax

  16. Camera Placement vs. Gaze Parallax • Ideally, cameras should be placed at the position of the eyes • Occlusion (either the camera or the image) • Cameras can be up to 5o before gaze parallax is perceived (Chen, 2002) [5°-7°]

  17. MultiView Directional Display • Big, Bright, High Resolution Display • Each view is provided by a projector • The projected image is reflected directly back in the direction of the projector • The image can be seen at varying heights only behind the projector

  18. Construction • Retroreflective Layer Reflects image back in direction of source • Vertical Diffuser Diffuses image vertically to accommodate varying viewing heights • Antireflective Sheet Reduces distracting glares due to glossy surface and front projection setup

  19. Brightness vs. Viewing Angle α -10O 0O

  20. MultiView Cameras Projectors MultiView Screen

  21. Cost • Small fixed cost • Variable costs increase linearly • Alternatives increase quadratically

  22. Easy to Join

  23. Talk Outline Motivation Prior Work Our Approach Experiment Discussion Conclusions

  24. Experimental Setup 3O

  25. Experimental Setup • 23 participants: UC Berkeley graduate and undergraduate students • Each Paid $10 • Experiments • Exp 1: 10 Trials (230 Total) • Exp 2: 10 Trials (230 Total) • Exp 3: 30 Trials (690 Total)

  26. Experiment 1 & 2:Mona Lisa Effect • Each researcher was asked to look or point at one of the five targets • Each participant was asked to circle which target each researcher appeared to be looking/pointing at • Repeated for 10 trials

  27. Experiment 1:Accuracy of Gaze Perception • Confusion Matrix • Each column representsactual target stimulus • Each Row represents perceived target • Ideally, this would be a diagonal matrix. • Participants were accurate in determining target of gaze • 90% were at most one target off

  28. Experiment 2:Accuracy of Gesture Perception • Confusion Matrix • Each column represents actual target stimulus • Each Row represents perceived target • Ideally, this would be a diagonal matrix. • Participants were accurate in determining target of gaze • 94% were at most one target off

  29. Experiment 3: Eye Contact • Researchers and Participants were paired off. • Researchers were asked to look at the eyes, at the cameras, below the eyes, to the left of the eyes or to the right of the eyes. • Each participant was asked if they felt like the partner was looking at them in the eyes.

  30. Mutual Gaze • No significant difference in positive responses between different gaze positions • “At Cam” yielded <100% positive response • Participants noted a strong sense of eye contact in the context of conversation Yes/No “Are they looking at me?”

  31. Virtual Viewing Distance Viewers sat 12’ from screen • projector throw distance Image needed to be scaled down to 66% • Scene:Screen size Resulting in a virtual viewing distance of 18’

  32. Future Work • New MultiView design iteration • 8’ Screen • Short throw projectors for closer virtual distance (from 18’ to 8’) • True Life Size Images

  33. Future Work:Shared Workspace VirtualObject Object Projector Glass Table

  34. Future Work • Using a new digital pipeline, we can control for aspects such as video quality and latency for low level experiments • New series of higher level experiments using new screen design • Trust formation • Turn Taking • Speech Patterns

  35. ? Conclusion • Standard video conference systems have spatial distortions which can adversely effect communication • Our perceptual experiments show that attention target cues can be accurately determined using our system • Our solution, MultiView, is the first practical video conferencing system to support spatially faithful group-to-group meetings

  36. Additional Slides

  37. MultiView David Nguyen Professor John Canny A Spatially Faithful Video-Conferencing System

  38. View From Left View From Right View From Center Position Dependant Views Participants are facing the right camera

  39. Current Technologies

  40. Functions of Gaze(Kendon, 1967) • Monitoring • “Is my partner listening to me?” • “Does my partner want to take over?” • Planning, Current Control, Checking • Regulating • Floor Control (Turn Taking) • Forestall/Demand a Response • Expressing – feelings or attitudes • Power, interest, point-granting, attention, agreement

  41. First Impressions: 7-38-55 Rule(Mehrabian, 1971) • In forming first impressions… • 7% is based on what you actually say • 38% is based on the way you say it • 55% is based on your appearance • Suggests that a lot is at risk in forming first impressions if visual nonverbal cues are not appropriately preserved

  42. Pop Quiz: First Impressions7-38-55 Rule(Mehrabian, 1971) In forming first impressions… • is based on content • is based on voice • is based on appearance Fill in the blank with [7%, 38%, or 55%] 7% 38% 55%

  43. The Video Tunnel • Variants Include… • Gaze-2 (Vertegaal et al., 1999, 2003) • GA Display (Monk and Gale, 2002) • The Teleprompter (Kahn, 1994) • Clearboard (Ishii and Kobayashi, 1992) • Reciprocal Video Tunnel (Buxton and Moran, 1990)

  44. Computer Vision/Graphics Approach(Gemmel et al., 2000)

  45. Human Accuity (Chen, 2002)

  46. Next Design Iteration • 8ft screen for life size images • Short throw projector for closer viewing distances (8’wide @ 9’6”) essentially halving virtual distance Ghosting • Multiple elements and uncontrollable diffusion causes ghosting effects

  47. Error by Viewing Position • In standard video conferencing systems, Mona Lisa Effect would cause greater error as viewing position deviates from “virtual camera position” • Finding: Viewing position had no significant effect on error • MultiView does not exhibit the Mona Lisa Effect

More Related