Date post: | 17-Jan-2017 |
Category: |
Education |
Upload: | stanfordcomputationalimaging |
View: | 503 times |
Download: | 3 times |
Slide 1
SID Display Week 2016Light Field, Focus-tunable, and Monovision Near-eye Displays
Gordon Wetzsteinwww.computationalimaging.orgMay 25, 2016
1
http://www.slideshare.net/StanfordComputationalImaging/light-field-focustunable-and-monovision-neareye-displays-for-virtual-and-augmented-realityDont take photos, just download the slides!
2
E&MWave Optics
Geometric Optics
Modern Signal Processingand OptimizationWhat is Computational Imaging?
3
++opticscomputationsensingComputational ImagingWhat is Computational Imaging?
[Devebec,Nayar,]HDR ImagingSuper-resolutionEDOFLight FieldsCompressive Imaging[Baker,Ben-Ezra,]
[Dowski,Nayar,][Levoy,][Baraniuk,]
4
What is Computational Imaging?
[Devebec,Nayar,]HDR ImagingSuper-resolutionEDOFLight FieldsCompressive Imaging[Baker,Ben-Ezra,]
[Dowski,Nayar,][Levoy,][Baraniuk,]
piximnexusphones
light.co
Ricoh
LytroGoogle??
5
computationoptics &electronicshuman visual system
++interactionComputational DisplaysWhat is Computational Imaging?[Seetzen,]HDR DisplaySuper-resolutionEDOFLight Fields
Near-eyeDisplays[Wetzstein, ]
6
Stanford Computational Imaging Group
=?
(Some) Technology ChallengesVergence-accommodation conflict (VAC)
Vestibular-visual conflict (motion sickness)
(Some) Technology ChallengesVergence-accommodation conflict (VAC)
Vestibular-visual conflict (motion sickness)
ARocclusionsaesthetics / form factorbattery lifeheatwireless operation
low-power computer visionregistration of physical / virtual world and eyes consistent lightingscanning real worldVAC more importantdisplay contrast & brightnessfast, embedded GPUs
Top ViewReal World:
Vergence & Accommodation Match!
17
VR/AR Display Optics = Magnifier
Top ViewNear-eye Displays Today (all stereo displays):
Vergence-Accommodation Mismatch!virtual image
19
Light Field Cameras
Light Field StereoscopeACM SIGGRAPH 2015
BacklightThin Spacer & 2nd panel (6mm)Magnifying LensesLCD PanelLight Field StereoscopeACM SIGGRAPH 2015
Target Light Field
Input: 4D light field for each eyeModel Courtesy of Bushmills Irish Whiskey
22
Multiplicative Two-layer Modulation
Model Courtesy of Bushmills Irish WhiskeyInput: 4D light field for each eye
23
Multiplicative Two-layer Modulation
Parallax over the Pupil
Model Courtesy of Bushmills Irish WhiskeyInput: 4D light field for each eye
24
Multiplicative Two-layer Modulation
Reconstruction:[Wetzstein et al 2012]Model Courtesy of Bushmills Irish WhiskeyInput: 4D light field for each eye
25
Traditional HMDs- No Focus CuesThe Light Field HMDStereoscopeLight Field StereoscopeACM SIGGRAPH 2015
26
Traditional HMDs- No Focus CuesThe Light Field HMDStereoscopeLight Field StereoscopeACM SIGGRAPH 2015
27
Model Courtesy of Paul H. ManningTraditional HMDs- No Focus CuesThe Light Field HMDStereoscopeLight Field StereoscopeACM SIGGRAPH 2015
28
Traditional HMDs- No Focus CuesThe Light Field HMDStereoscopeModel Courtesy of Paul H. ManningLight Field StereoscopeACM SIGGRAPH 2015
29
MonovisionACM SIGCHI 2016
30
MonovisionACM SIGCHI 2016
Evaluation through user studyDisplay ModesConventional
32
Evaluation through user studyDisplay ModesConventionalConventional + DoF
33
Evaluation through user studyDisplay ModesConventionalConventional + DoFAdaptive Focus
Gaussian Thin Lens Equation
o lens to screen distancedv lens to virtual image distf focal length of lens
34
Evaluation through user studyDisplay ModesConventionalConventional + DoFAdaptive FocusAdaptive Focus + DoF
35
Evaluation through user studyDisplay ModesConventionalConventional + DoFAdaptive FocusAdaptive Focus + DoFMonovision
36
User Preference study
User performance studies faster & more accurate
MonovisionACM SIGCHI 2016
37
User Preference study
User performance studies faster & more accurate
MonovisionACM SIGCHI 2016
38
Monovision
39
User Study SummaryMonovisionBetween conventional and adaptive focus in preferenceImproved time to fuse in visual clarity taskAdaptive focusMost preferred Improved time to fuse in visual clarity taskWith DoF improved accuracy in depth judgment taskSoftware only approach doesnt have measurable effect on VAC
40
Conclusion Use gaze contingent focus!MonovisionEasy to implement, effectiveGaze-contingent focusShows highest potential in our studyEye trackingFocus-tunable / actuated display
41
Conclusion Use gaze contingent focus!MonovisionEasy to implement, effectiveGaze-contingent focusShows highest potential in our studyEye trackingFocus-tunable / actuated display
42
EE 267 - http://stanford.edu/class/ee267/
Acknowledgements
Robert KonradFu-Chung HuangEmily Cooper
Gordon WetzsteinComputational Imaging GroupStanford Universitystanford.edu/~gordonwz
www.computationalimaging.org
45
User Preference Study
Asked users to visually explore 3 VR environmentsIn each one, uniquely rank the 5 display modesOverall quality of their experience46
User Preference - ResultsGoodBad
Bar heights represent the mean rankingError barsPoints represent individual sample points1 was the best (most preferred rank)Preferred the conventional mode the least47
User Preference - Results
Slightly better, but not statistically significant effectMean of 3.548
User Preference - Results
Mean of 2.2
49
User Preference - Results
Mean of 2.450
User Preference - Results
Mean of 3.2Falls somewhere in between adaptive, and conventionalInteresting because it isnt necessarily a natural effect with each eye seeing a differently focused imageAdaptive Focus (with and without DOF), showed statistical effect over the conventional modeFriedman test (and followup Tukey-Kramer method)51
Visual Clarity Study
Users identified the orientation of a small letter E placed on target within sceneMeasured reaction time (time-to-fuse) and accuracyTime-to-fuse correlated with amount of VACLarge VAC increases time to fuseExpected conventional to perform poorly compared to adaptive focusUser asked to fixate on initial target, at either 0.4, 1, or 2DSecond target appeared at relative distance: -0.5, -0.2, 0, 0.2, 0.5 DRemove relative object size cue22.59 arcminutes
52
Visual Clarity Reaction Time
fasterslower
Just showing closest case for nowInitial target at 2D with the secondary targets placed at the distances belowResults compared to conventional mode, representing by the dashed linePoints above the line were faster, and points below were slower53
Visual Clarity Reaction Time
fasterslower
Just adding in DoF effect didnt show a positive effect54
Visual Clarity Reaction Time
fasterslower
As expected adaptive focus reduces VAC and results in better reaction times55
Visual Clarity Reaction Time
fasterslower
Shows statistical significance compared to conventional mode in closer areas56
Visual Clarity Reaction Time
fasterslower
Shows statistical effect over conventional at closer target distancesThe relative increase at close target distances is largely due to the fact that reaction times in normal mode slowed down at these distances, while the the reaction times in monovision and adaptive focus modes remained largely stableAs expected, because the conflict for the conventional mode with the screen placed at 1.3m results in having the strongest conflict at close distances57
Visual Clarity Reaction Time
Not much effect in the farther viewing distances, as expected because the VAC is weaker there for the conventional mode58
Visual Clarity Accuracy
Average accuracy across all trial types 91%Not only better reaction times, but also better accuracy in close by targets59
Depth Judgement Study
Similar setup to visual clarity study with some initial target and relative distancesUsers asked to make relative depth judgement between two these two targetsExpected some improvement of adaptive focus over conventional because we support an additional depth cue (accommodation)Remove relative object size cue60
Depth Judgement Reaction Time
Did not observe a significant effectReaction times tended to slow down for closer targetsOne maybe think that monovision may be slightly slower because of reduction in stereopsis, but we didnt observe that effect61
Depth Judgement Accuracy
Adaptive Focus + DOF has statistical effect over conventional mode62