CHI 2015 Liwei Chan Chi-Hao Hsieh Yi-Ling Chen Shuo Yang - - PowerPoint PPT Presentation

▶
chi 2015
SMART_READER_LITE
LIVE PREVIEW

CHI 2015 Liwei Chan Chi-Hao Hsieh Yi-Ling Chen Shuo Yang - - PowerPoint PPT Presentation

Mar 05, 2023 573 likes •1.18k views

CHI 2015 Liwei Chan Chi-Hao Hsieh Yi-Ling Chen Shuo Yang Da-Yuan Huang Rong-Hao Liang Bing-Yu Chen National Taiwan University most effective interface: most effective interface: interface that we are trained to use at the

slide-1
SLIDE 1

CHI 2015

Chi-Hao Hsieh Yi-Ling Chen Shuo Yang Da-Yuan Huang Rong-Hao Liang Bing-Yu Chen

National Taiwan University

Liwei Chan

slide-2
SLIDE 2

most effective interface:

slide-3
SLIDE 3

most effective interface:

interface that we are trained to use at the longest

slide-4
SLIDE 4
slide-5
SLIDE 5
slide-6
SLIDE 6
slide-7
SLIDE 7

http://images.fineartamerica.com/images-medium-large/cyclops-kerri-ertman.jpg

Cyclops

slide-8
SLIDE 8

Full Body Interaction

slide-9
SLIDE 9

Full Body Interaction

slide-10
SLIDE 10

Related Work

slide-11
SLIDE 11

Kinect

slide-12
SLIDE 12

Kinect

Interaction zone

slide-13
SLIDE 13

On-body camera

wearable interaction zone

slide-14
SLIDE 14

[Digits, UIST ‘12]

slide-15
SLIDE 15

[OmniTouch, UIST ‘11]

slide-16
SLIDE 16

[ShoeSense, CHI‘12]

slide-17
SLIDE 17

a wearable network of 3-DOF inertial measurement units (IMU) for use in motion capture applications

MotionNode

h3p://www.mo*onnode.com/bus.html accelerometer

slide-18
SLIDE 18

Motion Capture Suit

slide-19
SLIDE 19

Motion Capture Suit

slide-20
SLIDE 20
slide-21
SLIDE 21

Cyclops

:: a single-piece wearable device that sees all.

slide-22
SLIDE 22

Cyclops

:: a single-piece wearable device that sees all.

slide-23
SLIDE 23

How wide the field-of-view

  • f the lens is required to see

the full body from users’ chest?

slide-24
SLIDE 24

GOPRO

slide-25
SLIDE 25

Feet?

GOPRO

slide-26
SLIDE 26

!!

slide-27
SLIDE 27

Camera’s field-of-view

How wide is wide enough?

slide-28
SLIDE 28

How wide is wide enough?

What visible to the camera

slide-29
SLIDE 29

How wide is wide enough to see the body like this?

slide-30
SLIDE 30
slide-31
SLIDE 31

235°

slide-32
SLIDE 32

a$ b$

9DOF IMU 235° Fisheye Lens Raspberry Pi IR LEDs

Proof-of-concept Prototype

slide-33
SLIDE 33
  • 1. Reorient the image to up right
slide-34
SLIDE 34
  • 2. Differentiate gesture types
slide-35
SLIDE 35

Wearable Forms

slide-36
SLIDE 36

Jump to 01:50

slide-37
SLIDE 37

Eco-Centric View of Body Gestures

slide-38
SLIDE 38

Pipeline of Body Gesture Recognition

Motion History Image Gesture Recognizer

Gesture Type Non-Gesture Motion Gesture

Foreground Extraction

?

slide-39
SLIDE 39

Pipeline of Body Gesture Recognition

Foreground Extraction Motion History Image Gesture Recognizer

Gesture Type Non-Gesture Motion Gesture

?

slide-40
SLIDE 40

Foreground Extraction

Region Filling Edge Detection Straighten to a strip

slide-41
SLIDE 41

Pipeline of Body Gesture Recognition

Foreground Extraction Motion History Image Gesture Recognizer

Gesture Type Non-Gesture Motion Gesture

?

slide-42
SLIDE 42

Motion History Image

:: an image template in which non-zero pixels simultaneously record the spatial and temporal aspects of motion.

slide-43
SLIDE 43

Motion History Image

Foreground MHI

slide-44
SLIDE 44

Pipeline of Body Gesture Recognition

Foreground Extraction Motion History Image Gesture Recognizer

Gesture Type Non-Gesture Motion Gesture

Foreground Extraction

?

slide-45
SLIDE 45

Pipeline of Body Gesture Recognition

Foreground Extraction Motion History Image Gesture Recognizer

Gesture Type Non-Gesture Motion Gesture

Foreground Extraction

Template Matching

slide-46
SLIDE 46

Experiment

slide-47
SLIDE 47

Experiment

1 2 3 4 5 6 7 8 9 10 11 12 13 15 14 16 17 19 20 18

Motion Exercise Stationary Exercise

20 Participants; their heights, weights, and BMI values are recorded.

iMHI

slide-48
SLIDE 48

Experiment

1 2 3 4 5 6 7 8 9 10 11 12 13 15 14 16 17 19 20 18

Motion Exercise Stationary Exercise

20 Participants; their heights, weights, and BMI values are recorded.

iMHI dMHI

slide-49
SLIDE 49

Experiment Result

Template Matching

slide-50
SLIDE 50

Random Decision Forest (RDF)

  • Data-driven learning algorithm
  • Notable example: Kinect
  • RDF: a set of decision trees;

each internal node is a weak learner

image

  • ffset intensity

image coordinate

  • ffset intensity

Feature response

f(I,x) = i( x + u ) - i( x + v)

slide-51
SLIDE 51

Experiment

RDF Template Matching

Random Decision Forest

slide-52
SLIDE 52

Experiment

RDF Template Matching

RDF$Layer$1$(IMU)$ RDF$Layer$2$(HMI)$

Determine gesture category Determine final gesture

Multi-Layer RDF

slide-53
SLIDE 53

Experiment with offset

30mm

slide-54
SLIDE 54

W/O IMU W/ IMU

TM +dMHI TM +iMHI RDF +dMHI RDF +iMHI 68.10% 59.80% 84.10% 80.40% 80.10% 71.20% 86.00% 86.40%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Standard Multi-Layered

Experiment with offset

slide-55
SLIDE 55

Applications

slide-56
SLIDE 56
slide-57
SLIDE 57

Discussion

  • Computer Vision Challenge
  • fisheye depth sensor
  • Social Acceptance by Gender
  • further design for female users
slide-58
SLIDE 58

Discussion

  • Computer Vision Challenge
  • fisheye depth sensor
  • Social Acceptance by Gender
  • further design for female users
slide-59
SLIDE 59
  • Cyclops: a single-piece wearable

device for full-body gesture input

  • The main contribution:

– the idea of determining body posture using an ego-centric perspective of the user.

  • We developed a proof-of-concept device to

demonstrate the feasibility of cycplos device.

Conclusion

slide-60
SLIDE 60

Thank you.

CHI 2013 UIST 2013 CHI 2015