3/23/2005 © Dr. Zachary Wartell 1 “Visual Cues for Perceiving Distances from Objects to...

3/23/2005 © Dr. Zachary Wartell 1

“Visual Cues for Perceiving Distances from Objects to Surfaces”

Helen H. Hu, Amy A. Gooch, Sarah H. Creem-Regehr, William B. Thompson

Presence, Vol. 11, No. 6, December 2002, 652– 664

Presentation: Revision 1.0


Introduction

• perception of impending contact between moving object and surface

• important during manipulation tasks• relevant visual range - “personal space” (Cutting

& Vishton)• paper presents two experiments in object-

surface distance perception:– 1) sub’s control movement of object – 2) sub’s just watch movement and report

object-surface distance• experiments IV: stereo, shadows, interreflections


Introduction (cont.)


Prior Work: Does stereo help? It depends…

• stereo not better than mono , [Kim87, Reinhart90, Barfield95]

• stereo → learn task more quickly [Drascic91]• stereo →performing task more quickly

[Spain90,Drascic 91, Yeh92,Hsu93,Ware96]• stereo →performing task more quickly (Cole90,

Barfield95]• generally stereo increases in effectiveness:

– as task is more difficult– as visual scenes has fewer other depth cues– for “personal space”


Prior Work: Shadows

• shadows important – [Yonas78,Kjelldahl95, Kersten 97, Madison 2001]

• shadow effectiveness – varies widely between tasks [Wanger92]– somewhat between individuals [Hu00]

• interaction of shadow & other depths cues– shadows sometimes override conflicting cues

[Bolj99]– shadows sometimes degrade task accuracy

and speed when added to stereo [Hubona99]


Prior Work: Interreflections

• in real-world they’re often visual indistinct (but in VR we can do anything!)

• evidence that they’re used perceptually [Kersten96]

• perhaps as spatial cues [Madison2001]– for contact perception interrefl.’s strong as

shadows


General Method

• shadows, interrefl.’s and stereo all computable at a cost, but stereo needs special hardware: So how do these compare in task performance?

• Equipment:– HMD: hi-res (1280x1024), 40.5 HOV– no head position tracking (no motion parallax)– no head orientation tracking (force sub to look

down in virtual world)– fixed IPD (6.5)


General Method (cont.)

• virtual environment: textured block approaching textured table

• IV: stereo vs biocular, shadows (on/off), interrefl. (on/off): 8 combinations

• table height and light position randomly varied between trials

• table texture adjusted for distance, so texture size doesn’t give distance info.!

• table height [46,60] cm below sub’s viewpoint• 5x5 cm block with infinite height• Table disappear/reappear between trials (no

jumping)


General Method (cont.)


Experiment I

• 6 sub’s• control block height with

physical block tracked via mech. tracker

• per sub.: 480 trials over 6 sessions

• pilot study indicates difficultswitching between stereo and biocular so:– turn-off HMD between sessions– vary combination of shadow & interrefl. in

session


Experiment 1 (cont.)

• sub’s have 1 s to bring block down and start back up (“bring to pt. just before contact”)– if contact, then give negative feedback &

discard trial (in exp. 22% trials discarded)


Exp. 1 Results


Relative vs Absolute Distance and Slope vs R2

• shadows and interrefl. are “scale-invariant”– only give relative dist., i.e. comparison of pair

of distances• stereo cable of:

– relative dist. – retinal disparity (relative to horopter)

– absolute dist. – if above combined with vergence

• human vis. system often assigns absolute dist. even given relative dist. cue

• implication: bad to compare slopes; rather compare R2


Results Table


Exp. 1 Result (cont.)


Result Table (cont.)

• all sub’s perform better with stereo than mono (higher R2)

• 2 x 2 x 2 (stereo x shadow x interrefl.) ANOVA indicates statistically significant effect of stereo but no other effects

• nonparametric test indicated statistically significant effect of shadows as well as stereo

• some sub’s show stat. sign. effect for shadows– 1 sub effect by shadow & not stereo

• across all sub’s interrefl. has no stat. sign. effect


Bias from negative feedback

• perhaps sub’s learned range of table heights from motor memory and stopped block at some average dist. above table even when visually they found dist. Ambiguous

• statistical analysis of data between sessions yields some stat. sign. learning effects– case BSI improves between session 1 & 2– case B improves between session 1 & 3

• bias may have been higher in cases with only rel. dist. cues


Experiment 2

• 6 sub’s: no glasses, tested for fusion, not from exp. 1

• block falls toward table and stops automatically at some distance above table

• sub: indicate the block-table dist. by sliding index finger & thumb on scale to match finger-thumb dist. to block-table dist.


Experiment 2

• IV: table-surface distance, DV: thumb-finger dist.

• sub’s perform task 48 times in biocular display – shadow and interrefl. combinations vary

• 1-3 weeks later: sub’s perform task 48 times in stereo display


Exp. 2 Results


Results: R2 and Cue Combinations


Discussion• Exp. 2 appears to have eliminated some biases

of exp. 1:– no cue case:

• Exp 1: sub’s perform better than random• Exp 2: sub’s perform closer to expected

• individually each sub.:– stat. sign. effect of stereo– stat. sign. effect of shadow under non-stereo

• some sub’s:– stat. sign. effect of interrefl. for non-stereo

• in pooled data all 3 occurred • interesting that abs. dist. cue (S) and rel. dist.

cue (BSI) yield similar performance


Conclusions

• two experiments over 8 combinations of stereo, shadow and interrefl.– stereo is strong cue for object-surface dist.– nonparametric statistics indicate shadows are

sign. in exp 1– exp 2. shows:

• case BSI similar performance to case S• shadows alone are effective but there are greater

individual differences

• individual diff’s:– perceived scaling ratios vary widely– some use shadows better; other’s interrefl.


Future Work

• generality of results needs further confirmation• study effect of task viewing and motion:

– exp 1. = dynamic, visual-motor task, closed-loop– exp 2. = open-loop matching task with dynamic and

stationary views• study effect of distance judgements along line-of-sight

versus perpendicular to LOS• study effect of varying geometry, surface markings, and

materials on effectiveness of shadow and interrefl.

3/23/2005 © Dr. Zachary Wartell 1 “Visual Cues for Perceiving Distances from Objects to...

Documents

Transcript of 3/23/2005 © Dr. Zachary Wartell 1 “Visual Cues for Perceiving Distances from Objects to...