Activity Recognition using Cell Phone Accelerometers

Activity Recognition using
Cell Phone
Accelerometers
Jennifer R. Kwapisz, Gary M. Weiss, Samuel A. Moore
Presenter : Ishara Amarasekera

Road Map
● Introduction
● Goal
● Contribution
● Data Collection
● Feature Generation & Data Transformation
● Experiment and Results
● Activity Recognition
● Conclusion and Future Work
1

Introduction
● Mobile devices are becoming increasingly
sophisticated.
● They are incorporating many diverse and powerful
sensors.
Accelerometer
Camera
Microphone
Compass
GPS
2

Introduction Cont.
Sensors :
● Location sensors (GPS)
● Audio sensors (i.e. Microphones)
● Image sensors (i.e. Cameras)
● Light sensors,
● Temperature sensors
● Direction sensors (i.e. Compasses)
● Acceleration sensors (i.e. Accelerometers)
3

Introduction Cont.
Because of,
● The small size of these smart mobile devices
● The substantial computing power
● The ability to send and receive data
● Their nearly ubiquitous use in our society
Mobile devices open up exciting new areas for
data mining research and data mining
applications
4

Goal
Describe and evaluate a system that uses
phone-based accelerometers to perform activity
recognition, a task which involves identifying
the physical activity a user is performing.
5

Contribution
● Collected data is served as a resource to other
researchers
● Demonstrate how raw time series accelerometer
data can be transformed into examples that can
be used by conventional classification algorithms
● Demonstrate that it is possible to perform
activity recognition with commonly available
(nearly ubiquitous) equipment and yet achieve
highly accurate results
● Bring attention to the opportunities available for
mining wireless sensor data and will stimulate
additional work in this area
6

Data Collection
● Through a smart phone with a simple GUI
● Record the user’s name, start and stop the data
collection
● Label the activity being performed
● Data was collected from 29 users
● Collected the accelerometer data every 50ms, so
there will be 20 samples per second
7

Feature Generation & Data
Transformation
● Raw time series data must be transformed into
examples as standard classification algorithms
cannot be directly applied to raw time-series
accelerometer data
● Data is divided into 10-second segments and
then generated features that were based on the
200 readings contained within each 10-second
segment.
● Duration of each segment is referred to as the
example duration (ED).
8

● 1 second --- 20 samples
● Divide data into 10 second segments – Example
Duration ()
10 s
20 X 10 = 200
Samples/ reading
10 s
10 s
ED ED ED
Transformation Cont.
9

● Generate informative
features based on 200 raw
accelerometer readings
● Each Reading contains X Y
Z values corresponding to
the three axes/dimensions
10

For this purposes,
● Z-Axis - Captures the
forward movement of the leg
● Y-Axis - Captures the upward
and downward motion
● X-axis captures the
horizontal movement of the
user’s leg
● These axes are relative to a
user
11

Generated 43 summary features which are variant
of six basic features
● Average [03]
● Standard Deviation [03]
● Average Absolute Difference [03]
● Average Resultant Acceleration [01]
● Time Between Peaks [03]
● Binned Distribution [30]
12

Preparing Data Set
● The resulting
examples contain 43
features and cover
twenty-nine users.
● This table shows the
percentage of the
total examples
associated with each
activity
Experiment and Results
13

Induce Data Mining Models
● Once the data set was prepared, the following
three classification techniques were used to
induce models for predicting the user activities
1. Decision trees (J48),
2. Logistic regression
3. Multilayer neural networks
● In each case the default settings were used
Experiment and Results Cont.
14

● This table specifies the
predictive accuracy
associated with each of
the activities, for each
of the three learning
algorithms and for a
simple “straw man”
strategy.
Summary Results of Activity Recognition Experiment
15

● For most common activities
such as Walking and Jogging
generally achieve accuracy
above 90%
● Results indicate that none of
the three learning algorithms
consistently performs best,
but the multilayer perceptron
does perform best overall.
In most cases high level of accuracy was obtained
16

Activity Recognition
● A periodic pattern is exhibited by
Walking
Jogging
Ascending and Descending stairs
● A distinct pattern is exhibited by
Sitting
Standing
17

Activity Recognition Cont.
Walking
● A series of high peaks for the y-axis, spaced out
at approximately ½ second intervals.
● The distance between the peaks of the z-axis and
y-axis data represent the time of one stride.
18

Jogging
● For jogging, similar trends are seen for the z-axis
and y-axis data, but the time between peaks is
less (~¼ second)
● The range of y-axis acceleration values for jogging
is greater than for walking, although the shift is
more noticeable in the negative direction.19

Descending Stairs
● Series of small peaks for y-axis acceleration that take place every ~½
second.
● Each small peak represents movement down a single stair.
● The z-axis values show a similar trend with negative acceleration,
reflecting the regular movement down each stair.
● The x-axis data show a series of semi-regular small peaks, with
acceleration vacillating again between positive and negative values
20

Ascending Stairs
● For ascending stairs, there are a series of regular peaks for
the z-axis data and y-axis data as well; these are spaced
approximately ~¾ seconds apart, reflecting the longer time
it takes to climb up stairs.
21

Sitting
● Sitting and standing do not exhibit any regular periodic
behavior and all of the acceleration values are relatively
constant
22

Standing
● Standing do not exhibit periodic behavior but do have
distinctive patterns, based on the relative magnitudes of the
x, y, and z, values.
23

Conclusion and Future Work
● Smart phone can be used to perform activity
recognition, simply by keeping it in ones pocket.
● Activity recognition can be highly accurate, with
most activities being recognized correctly over
90% of the time
● These activities can be recognized quickly, since
each example is generated from only 10 seconds
worth of data
● Can use activity recognition to implement some
interesting applications in the near future
24

Activity Recognition using Cell Phone Accelerometers

More Related Content

What's hot

Similar to Activity Recognition using Cell Phone Accelerometers

More from Ishara Amarasekera

Recently uploaded

Activity Recognition using Cell Phone Accelerometers