July 25, 2010 SensorKDD 2010 1 Activity Recognition Activity Recognition Using Using Cell Phone Cell Phone Accelerometers Accelerometers Jennifer Kwapisz, Gary Weiss, Samuel Jennifer Kwapisz, Gary Weiss, Samuel Moore Moore Department of Computer & Info. Department of Computer & Info. Science Science Fordham University Fordham University
25
Embed
July 25, 2010 SensorKDD 2010 1 Activity Recognition Using Cell Phone Accelerometers Jennifer Kwapisz, Gary Weiss, Samuel Moore Department of Computer &
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
July 25, 2010 SensorKDD 2010 1
Activity Recognition Activity Recognition Using Using
The ProblemThe Problem: use accelerometer : use accelerometer data to determine a user’s activitydata to determine a user’s activity
Activities include:Activities include: Walking and joggingWalking and jogging Sitting and standingSitting and standing Ascending and descending stairsAscending and descending stairs More activities to be added in future More activities to be added in future
workwork
July 25, 2010 4SensorKDD 2010
Applications of Activity Applications of Activity RecognitionRecognition
Health ApplicationsHealth Applications Generate activity profile to monitor overall Generate activity profile to monitor overall
type and quantity of activitytype and quantity of activity Parents can use it to monitor their childrenParents can use it to monitor their children Can be used to monitor the elderlyCan be used to monitor the elderly
Make the device context-sensitiveMake the device context-sensitive Cell phone sends all calls to voice mail when Cell phone sends all calls to voice mail when
joggingjogging Adjust music based on the activityAdjust music based on the activity
Broadcast (Facebook) your every activityBroadcast (Facebook) your every activity
July 25, 2010 5SensorKDD 2010
Our WISDM PlatformOur WISDM Platform
Platform based on Android cell phonesPlatform based on Android cell phones Android is Google’s open source mobile Android is Google’s open source mobile
computing OScomputing OS Easy to program, free, will have a large Easy to program, free, will have a large
market sharemarket share Unlike most other work on activity Unlike most other work on activity
recognition:recognition: No specialized equipmentNo specialized equipment Single device naturally placed on body (in Single device naturally placed on body (in
pocket)pocket)
July 25, 2010 6SensorKDD 2010
Our WISDM PlatformOur WISDM Platform
Current research was conducted off-lineCurrent research was conducted off-line Data was collected and later analyzed off-lineData was collected and later analyzed off-line
In future our platform will operate in In future our platform will operate in real-timereal-time
In June we released real-time sensor data In June we released real-time sensor data collection app to Android marketplacecollection app to Android marketplace Currently collects accelerometer and GPS Currently collects accelerometer and GPS
datadata
July 25, 2010 7SensorKDD 2010
AccelerometersAccelerometers
Included in most smart phones & other devicesIncluded in most smart phones & other devices All Android phones, iPhones, iPod Touches, etc.All Android phones, iPhones, iPod Touches, etc. Tri-axial accelerometers that measure 3 dimensionsTri-axial accelerometers that measure 3 dimensions
Initially included for screen rotation and Initially included for screen rotation and advanced game playadvanced game play
July 25, 2010 8SensorKDD 2010
Examples of Raw DataExamples of Raw Data
Next few slides show data for one user Next few slides show data for one user over a few seconds for various activitiesover a few seconds for various activities
Cell phone is in user’s pocketCell phone is in user’s pocket Earth’s gravity is registered as Earth’s gravity is registered as
accelerationacceleration Acceleration values relative to axes of Acceleration values relative to axes of
the device, not Earththe device, not Earth In theory we can correct this given that we In theory we can correct this given that we
can determine orientation of the devicecan determine orientation of the device
July 25, 2010 9SensorKDD 2010
StandingStanding
July 25, 2010 10SensorKDD 2010
SittingSitting
July 25, 2010 11SensorKDD 2010
WalkingWalking
July 25, 2010 12SensorKDD 2010
JoggingJogging
July 25, 2010 13SensorKDD 2010
Descending StairsDescending Stairs
July 25, 2010 14SensorKDD 2010
Ascending StairsAscending Stairs
July 25, 2010 15SensorKDD 2010
Data Collection Data Collection ProcedureProcedure
User’s move through a specific courseUser’s move through a specific course Perform various activities for specific timesPerform various activities for specific times Data collected using Android phonesData collected using Android phones Activities labeled using our Android appActivities labeled using our Android app
Data collection procedure approved by Data collection procedure approved by Fordham Institutional Review Board Fordham Institutional Review Board (IRB)(IRB)
Collected data from 29 usersCollected data from 29 users
July 25, 2010 16SensorKDD 2010
Data PreprocessingData Preprocessing Need to convert time series data into Need to convert time series data into
examplesexamples Use a 10 second example duration (i.e., Use a 10 second example duration (i.e.,
window)window) 3 acceleration values every 50 ms (600 total 3 acceleration values every 50 ms (600 total
values)values) Generate 43 total featuresGenerate 43 total features
Ave. acceleration each axis (3)Ave. acceleration each axis (3) Standard deviation each axis (3)Standard deviation each axis (3) Binned/histogram distribution for each axis (30)Binned/histogram distribution for each axis (30) Time between peaks (3)Time between peaks (3) Ave. resultant acceleration (1)Ave. resultant acceleration (1)
July 25, 2010 17SensorKDD 2010
Final Data SetFinal Data SetID Walk Jog Up Down Sit Stand Total 1 74 15 13 25 17 7 151 2 48 15 30 20 0 0 113 3 62 58 25 23 13 9 190 4 65 57 25 22 6 8 183 5 65 54 25 25 77 27 273 6 62 54 16 19 11 8 170 7 61 55 13 11 9 4 153 8 57 54 12 13 0 0 136 9 31 59 27 23 13 10 163
Utilized three WEKA learning Utilized three WEKA learning methods methods Decision Tree (J48)Decision Tree (J48) Logistic RegressionLogistic Regression Neural NetworkNeural Network
Results reported using 10-fold cross Results reported using 10-fold cross validationvalidation
July 25, 2010 19SensorKDD 2010
Summary ResultsSummary Results
July 25, 2010 20SensorKDD 2010
J48 Confusion MatrixJ48 Confusion Matrix
Predicted Class
Walk Jog Up Down Sit Stand
Actual
Class
Walk 1513 14 72 82 2 0
Jog 16 1275 16 12 1 1
Up 88 23 323 107 2 2
Down 99 13 92 258 1 2
Sit 4 0 2 3 270 3
Stand 4 1 2 7 1 208
July 25, 2010 21SensorKDD 2010
ConclusionsConclusions
Able to identify activities with good Able to identify activities with good accuracyaccuracy Hard to differentiate between Hard to differentiate between
ascending and descending stairs. To ascending and descending stairs. To limited degree also looks like walking.limited degree also looks like walking.
Can accomplish this with a cell phone Can accomplish this with a cell phone placed naturally in pocketplaced naturally in pocket
Accomplished with simple features and Accomplished with simple features and standard data mining methodsstandard data mining methods
July 25, 2010 22SensorKDD 2010
Related WorkRelated Work At least a dozen papers on activity recognition At least a dozen papers on activity recognition
using multiple sensors, mainly accelerometersusing multiple sensors, mainly accelerometers Typically studies only 10-20 usersTypically studies only 10-20 users
Activity recognition also done via computer Activity recognition also done via computer visionvision
Actigraphy uses devices to study movementActigraphy uses devices to study movement Used by psychologists to study sleep disorders, ADDUsed by psychologists to study sleep disorders, ADD
A few recent efforts use cell phonesA few recent efforts use cell phones Yang (2009) used Nokia N95 and 4 usersYang (2009) used Nokia N95 and 4 users Brezmes (2009) used Nokia N95 with real-time Brezmes (2009) used Nokia N95 with real-time
recognitionrecognition One model per user (requires labeled data from each user)One model per user (requires labeled data from each user)
July 25, 2010 23SensorKDD 2010
Future WorkFuture Work
Add more activities and usersAdd more activities and users Add more sophisticated featuresAdd more sophisticated features Try time-series based learning Try time-series based learning
methodsmethods Generate results in real timeGenerate results in real time Deploy higher level applications: Deploy higher level applications:
Same accelerometer data and same generated Same accelerometer data and same generated features but added 7 users (36 in total)features but added 7 users (36 in total)
If we group all of the test examples from one cell If we group all of the test examples from one cell phone and apply majority voting, achieve 100% phone and apply majority voting, achieve 100% accuracyaccuracy
Can be used for security or automatic Can be used for security or automatic personalizationpersonalization
Interested in GPS spatio-temporal data miningInterested in GPS spatio-temporal data mining
1 Kwapisz, Weiss, and Moore, Cell-Phone Based Biometric Identification, Proceedings of the IEEE 4th International Conference on Biometrics: Theory, Applications, and Systems (BTAS-10), September 2010.