christian holz patrick baudisch
DESCRIPTION
christian holz patrick baudisch. high-precision touch input based on fingerprint recognition. fachgebiet human-computer interaction. occlusion. fat finger. so touch is inaccurate or is it?. could it be that it is not the fingers but our touch devices that are wrong?. - PowerPoint PPT PresentationTRANSCRIPT
christian holzpatrick baudisch
high-precision touch inputbased on fingerprint recognition
fachgebiet human-computer interaction
occlusion
fat finger
so touch is inaccurate
or is it?
could it bethat it is not the fingers but our touch devices that are wrong?
Part 1 (science):even though screens are 2D, pointing is not
Part 2 (engineering):sensing fingers in 3D highly accurate touch
no fat fingerwe claim there is
problem
perceivedinstead, almost all observed targeting error comes from
probleminput point
perceived input point problem
target
[Benko, Wilson, & Baudisch 2006]
touch device perceives
offset
why we hope it’s the perceived input point problem?
the fat finger problem, in contrast is always noise = error
we can correct for it
offset
why we hope it’s the perceived input point problem?
the fat finger problem, in contrast is always noise = error
we can correct for it
while there is always an offset, we hypothesize thatthe offset depends on the pointing situation
our main hypothesis
so what does “pointing situation” mean?
!= [iPhone, Wang et al.]
1yaw
!=[Forlines et al., CHI’07]
2pitch
!=3roll
!=
4 finger shape
!=
4mental model
(… and there might be more e.g., head position/parallax…)
we ran auser studya non 2D-model
xy xy
current model
touch padscreen
nD xy
proposed model
touch padscreen
we ran auser studyuser study 1
task
1. target here
2. hit okay
task
1pad rotation (yaw)
90° 45° 15° 0° -15°
roll2roll
3pitch
90° 65° 45° 25° 15°
4user12 participants
(all students, so differencesamong them will be lower bound)
footswitch
on-screeninstructions
controlledhead position parallax
capacitivetouch pad
every trial recorded as a dot at the touch location
dependent
we measure targeting accuracy assuming perfect calibration size of ellipse that contains 95% of all samples.
example
7.5 m
m1.5 cm
main effects forroll, pitch, yaw, & participantID
hypotheses
2 pad rotations× 2 sessions (pitch, roll)× 5 angles× 6 repetitions per angle× 5 blocks
= 600 trials / participant
12 participants design
1 2 3 4 5 6
results
if the additional IVs had no impact,we would expect to see something like this
-15°0°
15°45°90°
rotate condition
no-rotate condition
but touch locations do indeed fall into clusters…
know
not
...
know
use
r
know
yaw
know
use
r...
know
3DO
F
know
use
r...0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
error bars are standard deviation
butto
n siz
e in
cm
for 9
5% a
ccur
acy results
requires 5.2mm button
~three times more accurate allow three times smaller device
trad
ition
alca
paci
tive
requires 15mm button
(participant #4, roll varied)
target
1cm
1pad rotation (yaw)
1pad rotation (yaw)
-15°0°
15°45°90°
rotate condition
no-rotate condition
(participant #4, roll only)2roll
1cm
3pitch 1025456590
all data by participant #1-6
1 2 3 4 5 6
tilt
roll4users
7 8 9 10 11 12
tilt
roll
all data by participant #7-124users
know
not
...
know
use
r
know
yaw
know
use
r...
know
3DO
F
know
use
r...0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
error bars are standard deviation
spre
ad in
cm
resultsrequires 5.2mm button
trad
ition
alca
paci
tive
requires 15mm button
how (in)accurate current devices are (button must be that big)
if we knew thepad orientation
if we knewfinger angles
also need to know user ID, or we will overcompensate for people like this one
shouldn’t we be able to make such a device?
Part 1 (science):even though screens are 2D, pointing is not
Part 2 (engineering):sensing fingers in 3D highly accurate touch
optical tracker
what do you mean: “not very practical”? retro reflective markers on finger… 6-16 camera setup…
makes a great “gold standard” implementation to test the concept
mobileok, maybe something a bit more
gets everything a traditional touchpad gets+ roll, pitch, yaw, & participantID
devices that sense touch and fingerprint already exist
this is very different from micro rolls [CHI 2009]
calibrationhave user touch a known target repeatedly and with different finger postures create database of (fingerprint, target offset)
useobtain fingerprint as user touches the devicelook up similar fingerprints in the databaseaggregate associated offsets (k nearest neighbor) and apply it
algorithm
user study 2
1 tracking device
fingerprintoptical tracker
“simulated capacitive” (just contact area)
2rotation
3 roll & pitch
roll -15° 0° 15° 45° 90°
pitch
15°
25°
45°
65°
90°
optical beats simulated capacitive by ~3x(based on user study 1)
fingerprint beats simulated capacitive(let’s find out by how much)
hypotheses
2 rotations× 13 angles× 5 repetitions per angle× 5 blocks
= 650 trials / participant
12 participants design
results
5.00
4.00
3.00
2.00
1.00
0.00
Error bars: +/- 1 SE
Mea
n spr
ead i
n mm
rawcapacitive
rotation-aware
capacitive
fingerprint-based
correction
tracker-based
correction
5.00
4.00
3.00
2.00
1.00
0.00
Error bars: +/- 1 SEM
ean s
prea
d in m
m
rawcapacitive
rotation-aware
capacitive
fingerprint-based
correction
tracker-based
correctionerror bars are standard deviation
spre
ad in
cm results
simulatedcapacitive
5.00
4.00
3.00
2.00
1.00
0.00
Error bars: +/- 1 SEM
ean
spre
ad in
mm
rawcapacitive
rotation-aware
capacitive
fingerprint-based
correction
tracker-based
correctionfingerprint optical
as expected a factor of 3x
works!
potential for improvement
conclusions
benefits1. make more reliable touch input devicesenter text on mobile touch device with high accuracy
2. avoid need for targeting aidssuch as offset cursor, shift, zooming,as they cost time and make touch less “direct”
3. make smaller mobile touch devicesbring touch input to watch-size mobile devices
use roll/pitch/yaw/userID touch device to
2/3 (7/8 of surface) of “fat finger problem”really stem from an oversimplified model of touch
touch is not 2D
model
find a closed representation of user profile speed up learning
combine with in-cell touch screens make small
next steps
fachgebiet human-computer interaction
thanks to my new group athasso plattner institutein berlin/potsdam
Christian HolzPh.D. Student, masters from Hasso Plattner InstituteMasters project with Steve Feiner at Columbia University, New York
.
Gerry Chuintern at Hasso Plattner InstituteMasters from U of Toronto
.
joe konstan: university of minnesotadaniel fisher: microsoft researchgary marsden: south africa, capetownmichael rohs: telekom labsscott klemmer: stanfordmark billinghurst: hitlab new zealandlucia terrenghi: vodaphone
come visit
fachgebiet human-computer interaction
open Ph.D./post doc position
all people
10 22 45 60 90
-10
0
10
45
90
without sense of rotation
10 22 45 60 90
-10
0
10
45
90
all people
with sense of rotation
0.40
0.30
0.20
0.10
0.00
Error bars: +/- 1 SE
raw capacitive rotation-awarecapacitive
per-anglecapacitive
Mea
n sp
read
in c
m
per user spread
Professor in computer science at Hasso Plattner Institute2002- research scientist at Microsoft Research, Redmond, WA2003- affiliate professor at University of Washington Seattle, WA2000-2002 research scientist at Xerox PARC2000 Ph.D. in computer science from TU Darmstadt
patrickbaudisch
Sean GustafsonPh.D. StudentMasters University of Manitoba, Canada on visualization, off-screen pointing
.
spatial cognitionon mobile