View
223
Download
0
Embed Size (px)
Citation preview
3/23/2005 © Dr. Zachary Wartell 1
“Dynamic Adjustment of Stereo Display Parameters”
Colin Ware, Cyril Gobrecht, and Mark Andrew Paton
IEEE Transactions on Systems, Man and Cybernetics—Part A: Systems and Humans, vol. 28, no. 1, January 1998, pg 56-65.
Presentation: Revision 1.0
3/23/2005 © Dr. Zachary Wartell 2
Introduction
• reviewed depth cues:– retinal disparity, diplopia, panum’s fusion area– occlusion– SFM– need to reduce/control screen parallax
• Warning: – [Ware] “virtual eye separation” = [Class] model eye
separation– doesn’t distinguish measurement in scaled space
• Accommodation/vergence conflict
3/23/2005 © Dr. Zachary Wartell 3
SFM: Motion Parallax
optic flow
3/23/2005 © Dr. Zachary Wartell 4
screen
wireframe cube(backlit)
SFM: Kinetic Depth Effect
3/23/2005 © Dr. Zachary Wartell 5
Modeled versus True Eye Separation
- v vs
v v v
SE zz
E z E S E z
3/23/2005 © Dr. Zachary Wartell 6
Experiment 1: Rate of Change of Eye Separation
Cyberscope
3/23/2005 © Dr. Zachary Wartell 7
Experiment 1: Rate of Change of Eye Separation
●plane infinite field of truncated pyramids moving towards viewer●plane tilted 70o from screen
3/23/2005 © Dr. Zachary Wartell 8
Varying Eye Separation
– IV: amplitude variation: 10%, 20%, 30%– IV: separations: 6.3, 4.2, 2.1 cm
Ex: 6.3 cm with 10% yields [5.67,6.3]
– 3 x 3 conditions– DV: sinusoidal variation frequency increased
until subjects noticed it
– within subjects, fully randomized– 9 subjects, each subject sees all conditions
3/23/2005 © Dr. Zachary Wartell 9
Results
-detectable frequency varies inversely with amplitude of oscillation-no significant interaction between amplitude and model eye sep.-worst case: average detectable frequency is 0.3 Hz occurs at max. amp. & max. e.s.
3/23/2005 © Dr. Zachary Wartell 10
Examine Peak Velocity
• significant effects of amplitude and model e.s. on Vp
• so peak velocity not primary determinant of response thresholds
• should keep rate of change of model e.s. < 0.2 cm/s
3/23/2005 © Dr. Zachary Wartell 11
Experiment 2
• subjects adjust model e.s. until diplopia occurs than back to comfortable value: find “max. comfortable value”
• Hypothesis: Sub’s will vary max. model e.s. based on scene depth
3/23/2005 © Dr. Zachary Wartell 12
Exp 2: Method
• vary scene depth via angle of “moving carpet”: 10 through 80 degrees
• sub’s make two e.s. settings per angle (diplopia & back to comfortable value)
• 12 subjects
• randomized,within-subjects
3/23/2005 © Dr. Zachary Wartell 13
Results
• considerable sub. variation• high neg. correlation (r2=0.99)
Ev = 18.5 – 0.149 θ• Implications:
drive algorithm to adjust e.s.; expect sub. variation
3/23/2005 © Dr. Zachary Wartell 14
Algorithm for Dynamic Disparity Adjustment
• two stages– cyclopean scale– modeled eye separation adjustment
• purpose: reduce a/c conflict and optimize stereo depth effectiveness
• use experimental result to control stereo adjustment
3/23/2005 © Dr. Zachary Wartell 15
Cyclopean Scale
• determine near point• scale about eye center to bring near point to
screen (equivalent of changing virtual e.s. in our term.)
3/23/2005 © Dr. Zachary Wartell 16
Dynamic Implementation
3/23/2005 © Dr. Zachary Wartell 17
Dynamic Implementation
PP'
Pl
Pr
3/23/2005 © Dr. Zachary Wartell 18
Adjustment of Eye Separation
• optional addition to cyclopean scale• model e.s. function from lower 95th percentile of
exp. 2 (model e.s. varies 14cm through 4 cm)
Ev = – 0.144 θ + 14.0,
θ = arctan (dz/dh)
Ev = – 0.144 arctan (dz/dh) + 14.0
3/23/2005 © Dr. Zachary Wartell 19
Adjustment of Eye Separation
• How to transform results from “magic carpet” to general scene?
Ev = -0.114 arctan (Zmax/Sh)+14.0
Zmax : zbuffer max Sh : screen height
Zmax
Sh θ
3/23/2005 © Dr. Zachary Wartell 20
Properties of Algorithm
• Zmax from 0 to ∞ → modeled e.s. 14 cm to 4 cm
3/23/2005 © Dr. Zachary Wartell 21
Properties of Algorithm
• relate max. displayed depth after m.e.s. adjustment to max. depth before (Zmax )
• assume modeled depth is infinite: maxv
a v
SEZ
E E
3/23/2005 © Dr. Zachary Wartell 22
Individual Differences
• customize with MaxSep
(Ev = -0.114 arctan (Zmax/Sh)+14.0)
max4.0arctan
90.0vh
ZMaxSepE MaxSep
S
3/23/2005 © Dr. Zachary Wartell 23
Controlling Rate of Change
• Options:• leave m.e.s. uncontrolled• use running average last 5 of model e.s. to smooth
change• use threshold 0.2 cm/s for m.e.s.
• problem: sudden changes in scene depth cause diplopia for some period
• Tradeoff: anecdotally its better to allow abrupt m.e.s. change than allow diplopia
3/23/2005 © Dr. Zachary Wartell 24
Experiment 3: Distortion with Changes in M.E.S.
• investigate perceived distortion: How much do users notice distortion?How do users rate it compared to diplopia?
• Method:– terrain map, 80 x 80 grid height field
colored by height, shaded, shadows
– pre-set flyby path: downlooking flying in, rotate to horizontal view, flyout
– Goal: large changes in view distance and relative depths
3/23/2005 © Dr. Zachary Wartell 25
6 Conditions
• no algorithm – m.e.s. = 6.4 cm• cyclopean scale only• dynamic adjustment – max m.e.s. 6.4 cm• dynamic adjustment – max m.e.s. 12.8 cm• dynamic adjustment – max m.e.s. 19.2 cm• dynamic adjustment – max m.e.s. 25.6 cm
3/23/2005 © Dr. Zachary Wartell 26
Measurements
• sub’s report perception of double images and note “overall shape of surface appeared to change”
• Rate distortion (0-4)• Rate eye strain (0-4)
• 7 subjects, within subjects, randomized presentation of all cond.’s
3/23/2005 © Dr. Zachary Wartell 27
Results: Diplopia
• “no algorithm” – all sub’s report double images when scene was distant
• “dynamic adj. with max 25 cm” – 2 / 7 reported double images
• indicates support for algorithm’s ability to control diplopia
3/23/2005 © Dr. Zachary Wartell 28
Results: Distortion
• See chart• one report of eye strain but exp. Only lasted 20
minutes
3/23/2005 © Dr. Zachary Wartell 29
Conclusions
• dynamic disparity adjustment used in terrain vis. app. algorithm provides;– strong depth cues & minimize diplopia
• compare Williams & Parrish: – old alg. always maps modeled space to fixed
depth range in display space– new alg. limits exaggeration of shallow scenes
(factor of 2)• SFM cues still dominate in giving sense of
“flying”
3/23/2005 © Dr. Zachary Wartell 30
Future Work
• investigate penalty from distortion when depth judgements are important
• improve depth buffer sampling; small # of sample points:– works best for large extended surfaces– missing small/narrow features
• alternative: use bounding boxes, but more complex implemention
3/23/2005 © Dr. Zachary Wartell 31
Questions, Comments
• long term (> 20 min.) study of eye strain affect(subjective questionnaire versus A/V ratio test)
• user study comparing Will. and Parrish. technique• why test angle of magic carpet? not direct depth?• better way to rate distortion (tease out diplopia effect in
cond. #1)• use near point from previous frame & scale about
cyclopean eye• battery of tests for best way to get zbuffer info.—GPU
solution?