16
Feature Extractors for Integration of Cameras and Sensors during End-User Programming of Assistive Monitoring Systems Alex Edgcomb Frank Vahid University of California, Riverside Department of Computer Science 1 of 16 ? Motion sensor

Alex Edgcomb Frank Vahid University of California, Riverside Department of Computer Science

  • Upload
    hina

  • View
    37

  • Download
    0

Embed Size (px)

DESCRIPTION

Feature Extractors for Integration of Cameras and Sensors during End-User Programming of Assistive Monitoring Systems. ?. Alex Edgcomb Frank Vahid University of California, Riverside Department of Computer Science. Motion sensor. Sensors and actuators in MNFL [1] for end-user programming. - PowerPoint PPT Presentation

Citation preview

Page 1: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Feature Extractors for Integration of Cameras and Sensors during

End-User Programming of Assistive Monitoring Systems

Alex EdgcombFrank Vahid

University of California, RiversideDepartment of Computer Science

1 of 16

?Motion sensor

Page 2: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Sensors and actuators in MNFL [1] for end-user programming

Alex Edgcomb, UC Riverside 2 of 16

“Person at door”

LED lights in house

“Person at door”

Outdoor motion sensor

Doorbell

• Assistive monitoring• User customizability essential [2][3]

[1] Edgcomb, A. and F. Vahid. MNFL: The Monitoring and Notification Flow Language for Assistive Monitoring. Proceedings 2nd ACM International Health Informatics Symposium, 2012. Miami, Florida.[2] Philips, B. and H. Zhao. Predictors of Assistive Technology Abandonment. Assistive Technology, Vol. 5.1, 1993, pp. 36-45.[3] Riemer-Reiss, M. Assistive Technology Discontinuance. Technology and Persons with Disabilities Conference, 2000.

Page 3: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Porch light

LED lights in house

Expanding the previous example

Alex Edgcomb, UC Riverside 3 of 16

“Person at door”

“Person at door”

Outdoor motion sensor

Doorbell

Light sensor

Page 4: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Webcams are cheap

4 of 16Alex Edgcomb, UC Riverside

Page 5: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Webcams can do more than sensors

Fall down at home

In room for extended time

Can do same as some sensors

Motion sensorLight sensor

5 of 16Alex Edgcomb, UC Riverside

Identify personat front door

Page 6: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Problem: Integration of webcams and sensors

6 of 16

Homesite

Commercial approach:

Alex Edgcomb, UC Riverside

?

Outdoor motion sensor

Page 7: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Solution: Feature extractor

7 of 16

92Integer stream

output

0

100

Alex Edgcomb, UC Riverside

Extract some feature

Video stream input

Page 8: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Identify person at door in MNFL

Alex Edgcomb, UC Riverside 8 of 16

Outdoor motion sensor

Page 9: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Person in room for extended period of time in MNFL

9 of 16Video’s YouTube link

Alex Edgcomb, UC Riverside

Page 10: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Many feature extractors are possible

10 of 16Alex Edgcomb, UC Riverside

Page 11: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Are feature extractors usable by lay people? Two usability trials.

• 51 participants• Trials required as 1st lab assignment• Non-engineering/non-science students at UCR

1 2 3 4 50

10

20

30

40

50

60

70

80

90

100

Programming experience(1- little/none; 5- a lot)

Perc

enta

ge o

f par

ticip

ants

11 of 16Alex Edgcomb, UC Riverside

Page 12: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Participant reference materials

• One-minute video showing how to spawn and connect blocks.

• Overview picture

12 of 16Alex Edgcomb, UC Riverside

Page 13: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Example challenge problem

13 of 16Alex Edgcomb, UC Riverside

actual participant solution

Page 14: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Trial 1: Increasingly challenging feature extractor problems

25 participants14 of 16

10 9 7-8 5-6 3-4 1-2 00%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Fall detector (easy) Visitor detector (medium)

Leave at night detector (hard)

Rubric score

Perc

ent o

f sub

ject

s

Alex Edgcomb, UC Riverside

Page 15: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Trial 2: Feature extractor vs logic block

26 participants15 of 16Alex Edgcomb, UC Riverside

10 9 7-8 5-6 3-4 1-2 00%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Visitor detector (feature)

Emergency button (logic)

Rubric score

Perc

enta

ge o

f sub

ject

s

Page 16: Alex  Edgcomb Frank  Vahid University of California, Riverside Department of Computer Science

Conclusions

• Feature extractors– Elegant integration of cameras and sensors– Quickly learnable by lay people

• Future work– Develop additional feature extractor blocks– Trade-off analysis between privacy,

communication, and computation16 o f 16Alex Edgcomb, UC Riverside