You are on page 1of 22

Indoor vs.

Outdoor Depth Perception for


Mobile Augmented Reality
Mark A. Livingston1, Zhuming Ai1, J. Edward Swan II2, Harvey S. Smallman3

1Naval Research Laboratory


2Mississippi State University

3Pacific Science & Engineering


Depth Perception in AR
• Typical VE depth perception is compressed
• AR depth perception has been observed to be:
– Overestimated at short distances [Rolland95]
– Underestimated from occluded locations at short distances [Ellis88]
– Underestimated at medium distances; overestimated at far [Swan06]
– Underestimated at medium distances with many protocols [Swan07]
• Real depth perception has been observed to be:
– Similar to AR depth perception [Livingston05]
– Unaffected by indoor/outdoor environment change
[Lappin06,Bodenheimer07]

2
Goals for Current Experiment

1. Compare indoor and outdoor AR depth perception


at medium- and far-field distances
2. Determine whether strong linear perspective cues
of indoor environment could be replicated to
improve outdoor performance
3
User Study Design
Target
• Task: match the depth of
virtual target to a real
referent
– Use “infinite” trackball to
control depth
• Twelve subjects
– Eight men, four women
– Age { 22 – 51 }; mean=35.6 Red
– All passed stereo acuity test
(at least 7/9)
Referents

nVisorST
4
User Study Design
• Independent Variables
– Environment { indoor, outdoor } within subjects
– Tramlines { on, off } within subjects
– Gridpoints { on, off } within subjects
– Distance within subjects
• { 4.83, 9.66, 14.49, 19.32, 24.15, 28.98, 33.81, 38.64 } meters
– Repetition within subjects
• { 1, 2, 3, 4, 5 }

• Dependent Variables
– Signed, absolute, or normalized error
– Response time
– User position
– NASA TLX after each environment

• Design Details
– Seven to fourteen days between environments – not counterbalanced
– Colors were presented in random permutation; target at random start position
– Yields 160 trials per environment per subject; 3840 total data points 5
User Study Design
Gridpoints on
Tramlines on

6
Hypotheses

• Outdoor environment would have greater


error due to loss of perspective
• Tramlines and gridpoints would reduce the
error in depth matching
• Increasing error and decreasing precision
with increasing distance
• As seen in previous study, improved
performance with repetition
– Despite the lack of feedback
7
Main Effect:
Environment on Normalized Error
• Underestimated
depth in indoor
environment;
overestimated
depth in outdoor
environment
• Not an effect of
distance, since this
is normalized error F(1,11)=11.032, p=0.007

8
Main Effect:
Repetition on Normalized Error
• Users increased
their estimate with
increasing repetition
• No feedback provided
• Not a reduction in the
error for the outdoor
environment; did
represent a reduction
of error for the indoor F(4,44)=6.613, p=0.000
environment

9
Main Effect:
Gridpoints on Signed Error
• Not a “real” effect;
entirely due to
sixth referent
• Disappointing;
implies that
gridpoints were
not truly useful

F(1,11)=5.279, p=0.042

10
Main Effect:
Distance on Absolute Error
• Absolute error
grew and precision
decreased with
increasing target
distance
– Consistent with
general results on
depth perception
– Indicates validity
of the task F(7,77)=13.391, p=0.000

• Difference between
indoor and outdoor rates
11
Main Effect:
Distance on Response Time
• Typical effect:
Users were
slower for more
distant targets
• Uniform random
variable for initial
position should
have evened out
effect of distance; forced F(7,77)=25.620, p=0.000
initial position to be at least
five meters from correct position
12
Main Effect:
Repetition on Response Time
• Users got faster with
repetition
– Likely due to
familiarity with
equipment and
order of colors
– Fourth and fifth
repetitions nearly
equal speed
• Does not imply that there F(4,44)=16.717, p=0.000

were no mistakes in color


identification; outliers do exist
13
Effect of Tramlines
• Interaction found
between tramlines
and distance
• For three most
distant referents,
tramlines helped
outdoors
• These referents
were beyond the
range of stereo as a depth cue
Restricted analysis
• At this distance, our previous study showed F(1,11)=5.356,
a change to overestimation from p=0.041
underestimation
14
Outlier Removal

• Some subjects
were not very
good at the
task
• Based on the
graph of error
by subject, we
re-ran the
analysis with subjects
S01 (low), S03 (high), and S12 (high) removed
• Main effects remained; new interaction for tramlines
15
Subjective Results

• Users completed NASA TLX after indoor session


and after outdoor session
NASA TLX
Environment Sequence Mean Range
Indoor First 52 35 – 69
Outdoor +7 to +14 days 44 19 – 77

• Most users thought they did better outdoors (10-2)


• Slight increases in eye strain, fatigue, dizziness, and
pain during both and indoor and outdoor sessions
16
Possible Error Models
• Ponzo illusion
– Predicts change in perceptual size
– May thus cause stretch of perceived depth,
but should have caused this indoors too
• Brightness differences
– Predicts compression for distant referents, stretch for nearby referents
– Colors faded during outdoor sessions
• Height and Declination
– Amount of variation was 18cm maximum Typo in proceedings (says 28)
– Kept angle of declination constant by using platforms under referents
– Predicts stretch for nearby, compression for distant referents
d Δd
• Constant perceptual size -18.3°
h (h+Δh)
– Predicts compression
17
Summary: Results

• Observed a difference for AR depth perception between indoor


(underestimation) and outdoor (overestimation)
– Indoor data does not agree with our own previous studies which showed both
underestimation (medium-field) and overestimation (medium-to-far-field)
• Appears that tramlines helped users correct their depth estimates
somewhat for distant targets in the outdoor environment
23
Future Work

• Counterbalance indoor and outdoor sessions


– Order effects have been noticed with depth
perception in real and immersive environments
– These studies used different protocols for
estimation and did not study AR
• Investigate difference between this data and
our previously-collected data
• Effect of registration?

24
Future Work

• Outlier definition and analysis

Is this an outlier?
This is an outlier

25
Acknowledgments

• Anonymous subjects, anonymous reviewers


• Support provided by NRL Base Program

Thank you for your attention!

26
Indoor vs. Outdoor Depth Perception for
Mobile Augmented Reality
Mark A. Livingston, Zhuming Ai, J. Edward Swan II, Harvey S. Smallman

Any Questions?

http://www.ait.nrl.navy.mil/3dvmel/
27

You might also like