Probabilistic Palm Rejection Using Spatiotemporal Touch Features - - PowerPoint PPT Presentation

▶
probabilistic palm rejection using spatiotemporal touch
SMART_READER_LITE
LIVE PREVIEW

Probabilistic Palm Rejection Using Spatiotemporal Touch Features - - PowerPoint PPT Presentation

Feb 01, 2024 216 likes •591 views

Probabilistic Palm Rejection Using Spatiotemporal Touch Features and Iterative Classification Julia Schwarz, Robert Xiao, Jennifer Mankoff, Scott E. Hudson, Chris Harrison ? ? ? ? pen palm palm palm Prior Software-Only Approaches

slide-1
SLIDE 1

Probabilistic Palm Rejection Using Spatiotemporal Touch Features and Iterative Classification

Julia Schwarz, Robert Xiao, Jennifer Mankoff, Scott E. Hudson, Chris Harrison

slide-2
SLIDE 2
slide-3
SLIDE 3
slide-4
SLIDE 4

? ? ? ?

slide-5
SLIDE 5

palm palm palm pen

slide-6
SLIDE 6

Prior Software-Only Approaches

slide-7
SLIDE 7

Ewerling et. al, ITS ‘12

slide-8
SLIDE 8

palm rejection region

slide-9
SLIDE 9

Vogel et al. CHI ‘09

slide-10
SLIDE 10

Penultimate for iOS Bamboo Paper for iOS

slide-11
SLIDE 11

Our Approach

Collection of decision trees, spatiotemporal features. Handedness and orientation agnostic. No calibration required.

slide-12
SLIDE 12
slide-13
SLIDE 13

green = stylus blue = palm Palms have large radius. Palms flicker in and out. Stylus is isolated. Palms move little, styluses have 
 smooth trajectories.

slide-14
SLIDE 14
slide-15
SLIDE 15

t = 0

slide-16
SLIDE 16

Instantaneous Features Touch radius Distance to other touches on screen t = 0

slide-17
SLIDE 17

t = 0 t = 5ms t = 10ms

Touch Sequence Features [µ,σ, min, max] touch radius over sequence [µ,σ, min, max] distance to other touches in sequence [µ,σ, min, max] velocity, acceleration

slide-18
SLIDE 18

t = 0 t = 5ms t = 10ms t = -10ms

Touch Sequence Features [µ,σ, min, max] touch radius over sequence [µ,σ, min, max] distance to other touches in sequence [µ,σ, min, max] velocity, acceleration

slide-19
SLIDE 19

* leftmost point is at t = 1ms

train: 11,000 instances from 3 people test: 11,000 instances from 2 different people

train and test data gathered in different locations and on different days

slide-20
SLIDE 20

Window size of ~250ms would be ideal. Want to provide immediate feedback to the user.

slide-21
SLIDE 21

t 0ms 50ms 100ms …

  • 50ms
  • 100ms
slide-22
SLIDE 22

t 0ms = palm 50ms 100ms …

  • 50ms
  • 100ms
slide-23
SLIDE 23

t 0ms = palm 50ms 100ms …

  • 50ms
  • 100ms

= stylus

slide-24
SLIDE 24

t 0ms = palm 50ms 100ms …

  • 50ms
  • 100ms

= stylus = stylus

slide-25
SLIDE 25

t 0ms = palm 50ms 100ms …

  • 50ms
  • 100ms

= stylus = stylus

= stylus

final classification

slide-26
SLIDE 26

Demo

slide-27
SLIDE 27

Evaluation

Penultimate vs. vs. Bamboo Our App

slide-28
SLIDE 28

symbols:

slide-29
SLIDE 29

symbols: false negative

slide-30
SLIDE 30

% pen strokes classified as pen strokes

error bars = 95% confidence interval

slide-31
SLIDE 31

symbols: false positive

slide-32
SLIDE 32

palm accuracy

# of palm ‘splotches’ per pen stroke

*error bars = 95% confidence interval

slide-33
SLIDE 33

Takeaways

Waiting to see how sensed input evolves before making a decision improves recognition accuracy. Need a system that can show immediate feedback, but that can refine the interface as more information is presented.

slide-34
SLIDE 34

Thank you!

julia@qeexo.com Special thanks to Jim Baur for photography assistance

  • Also, thank you to our sponsors:
slide-35
SLIDE 35

Why a decision tree?

slide-36
SLIDE 36

Limitations

No multitouch gestures (yet) Algorithm overly reliant on touch radius Accuracy hit of 1% when not using radius features Difficult to implement on platforms that do not expose touch radius