ENGLISH 7_Q4_LESSON 2_ Employing a Variety of Strategies for Effective Interp...
ExposureSense Demo
1. EXPOSURESENSE: INTEGRATING
DAILY ACTIVITIES WITH AIR
QUALITY USING MOBILE
PARTICIPATORY SENSING
Bratislav Predic*, Zhixian Yan† , Julien
Eberle‡ , Dragan Stojanovic*, Karl Aberer‡
* University of Nis, Serbia
† Samsung Research, USA
‡ EPFL, Switzerland
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
2. Sensors and smartphones
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
2
Modern smartphones accompany users 24/7
and have
Increasing number of integrated sensors
(accelerometer, gyroscope, sound/light sensor,
camera, compas,...)
Continuously increasing processing and storage
capacity
Powerfull sensor platforms
Sensor commonly used in research
Accelerometer : detecting user activity
accept/reject call, initiate file transfer, snooze
alarm…
Using data mining techniques to infer more
complex user physical activities
3. Smartphones and air quality
monitoring
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
3
Air quality monitoring – traditional approach
Fixed or mobile sensing nodes
OpenSense project in Switzerland
- sensors on top of public transport vehicles
Smartphones and pollution sensing
Integrated audio analysis as noise pollution
indicator
Beyond embedded sensors
USB pluggable air quality sensors (ozone O3
sensor)
4. Activity/air quality correlation
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
4
User’s activity and air quality measurements
Usually treated as fairly independent
ExposureSense
Correlation of activity and air quality data
Bridges the gap and estimates user’s exposure to
air pollution
Combination of air quality sensing modes
PM10 monitoring stations on public transport
vehicles
Pluggable O3 sensor for smartphones
5. ExposureSense
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
5
Provides additional knowledge from correlation of
data from different sensors
Technical challenges
Develop uniform interface for
sensor access
Important for “virtual” sensors
capturing phone states
Sensor adapter/wrapper
middle layer
6. System architecture
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
6
The abstraction layer was implemented using
sensor probe approach and software
components of the Funf – open sensing
framework
Main components
User activities recognition
Acquisition of air quality from pluggable sensors
Acquisition of air quality from external sensor
network
Daily exposure estimation
Mobile front-end interface
7. User activities recognition
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
7
Implemented by extended
accelerometer probe
Encapsulates inference engine for
activity recognition through:
Sampling
Extracting features
Building classification model
Classifying unknown accelerometer streams
J48 classification decision tree used in
experiments
8. User activities recognition
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
8
Accelerometer data
features used
Mean value
Standard deviation
Correlation
Acceleration vector
intensity mean value
Energy
Entropy
Time and frequency
domain
9. User activities recognition
engine
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
9
Features calculated per
accelerometer axis in
time and frequency
domain
J48 decision tree classifier
||
||
||
1
i ix
Energy
n
i
ii xpxpEntropy
1
2 )(log)(
10. Acquisition of air quality from
pluggable sensors
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
10
Participatory air quality sensing
complements external sensor
networks
Example: OpenSense deploys rich
set of
air quality sensors on top of public
transport vehicles
Smartphones can act as both
consumer and contributor to sensing
network
As a client of Global Sensor Network
11. Acquisition of air quality from
external sensor network
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
11
Smartphone acts as a contributor to
Global Sensor Network (GSN)
Current smartphones lack
integrated air
quality sensors
USB pluggable sensor platform
Local storage and publish data to
GSN
Interpolation with external sensor
nodes
network data to estimate exposure
12. Daily exposure estimation
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
12
Correlating inferred activity data with air quality
data acquired from pluggable and external
sensors
Exposure intensity is estimated based on
activity type detected and burned calories per
acitivity according to MET research
13. Mobile client front-end
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
13
Tabbed view system
Raw sensor data view
Frequency domain
accelerometer data view
Activity/air quality
timeline data view
Map data view
Android broadcast
communication mechanism
Service front-end
14. Mobile client demonstration
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
14
Calories burn MET based
calculation
Activity history on timeline and
map view
CB = (BMR/24) * MET * T
BMRmale = (13.75*WKG) + (5*HC) - 6.76*AGE + 66
BMRfemale = (9.56*WKG) + (1.85*HC) - 4.68*AGE +655
CB - calories burnt
BMR - basic metabological rate
WKG - weight in kg
HC - height in cm
T - time in h
15. Mobile client demonstration
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
15
Air quality parameter chooser
Timeline view of chosen air
quality parameter
Air quality readings map view
Diary-type calendar history
overview
Daily activity and estimated
exposure
16. Video demonstrations
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
16
ExposureSense Android Client
Demo
17. Conclusion and future work
MOVE COST Final Conference, Technical University of Vienna, September 30th – October 1st, 2013.
17
Personalized daily diary integrating user activities
and air quality
A building block for next generation personalized
healthcare applications based on smartphones
Future research directions
Analysis and mining of stored data about user
activities, calories burnt and pollution exposure,
detecting interesting patterns, providing
recommendations
Integrate more sensor inputs and virtual sensors: user
interaction, profile, social network activities, etc.
Notes de l'éditeur
Presentation slide for courses, classes, lectures et al.