ORIGINAL RESEARCH ARTICLE
published: 07 November 2012
Using mobile phones for activity recognition in
MarkV.Albert1,2*, SantiagoToledo2, Mark Shapiro1,2and Konrad Kording1,2
1Sensory Motor Performance Program, Rehabilitation Institute of Chicago, Chicago, IL, USA
2Department of Physical Medicine and Rehabilitation, Northwestern University, Chicago, IL, USA
Ryuji Kaji,Tokushima University
Pedro Chana, Universidad de Santiago
de Chile, Chile
Benzi Kluger, University of Colorado
Maria Fiorella Contarino, Academic
Medical Center, Netherlands
Mark V. Albert, Sensory Motor
Performance Program, Rehabilitation
Institute of Chicago, 345 E Superior
Street room 1479, Chicago, IL 60611,
everyday movements and classify those movements into activities. Using accelerometer
data we estimate the following activities of 18 healthy subjects and eight patients with
Parkinson’s disease: walking, standing, sitting, holding, or not wearing the phone. We
use standard machine learning classifiers (support vector machines, regularized logistic
regression) to automatically select, weigh, and combine a large set of standard features
for time series analysis. Using cross validation across all samples we are able to correctly
identify 96.1% of the activities of healthy subjects and 92.2% of the activities of Parkin-
son’s patients. However, when applying the classification parameters derived from the
set of healthy subjects to Parkinson’s patients, the percent correct lowers to 60.3%, due
to different characteristics of movement. For a fairer comparison across populations we
also applied subject-wise cross validation, identifying healthy subject activities with 86.0%
accuracy and 75.1% accuracy for patients.We discuss the key differences between these
populations, and why algorithms designed for and trained with healthy subject data are not
reliable for activity recognition in populations with motor disabilities.
Keywords: mobile phone, accelerometer, activity recognition, Parkinson’s disease
has the potential to better inform patient care.With more precise,
objective measures treatment alternatives can be evaluated more
definitively. This is particularly important in motor disabilities
such as Parkinson’s disease (PD) that respond to an increasing
variety of treatment options,including drugs and various exercise
therapies (Palmer et al., 1986; Schenkman et al., 1998; Goodwin
et al., 2008; Dibble et al., 2009). Quantifying symptoms both in
the clinic and at home has the potential to provide functional
measures better associated with quality of life (Ellis et al., 2011).
ical evaluations require a patient to travel to the location of a
health care provider, and testing is expensive in terms of money
per week during research studies. For better temporal resolution,
ically indicate their activities suffers from a number of problems.
First, it is subjective, leading to changes based on the mental state
of the subject. Also, because of the inconvenience to patients, it
is difficult to achieve high compliance, with one study indicating
only an 11% compliance rate using journaling (Stone et al.,2003).
It would be helpful to develop a measure of evaluating patient
mobility that is both frequent and convenient for the subject.
can cost as little as one dollar, and can measure movement in any
direction as well as orientation relative to gravity. For example,by
of time running and walking based on the periodic movements.
these accelerometers at specific locations on the body – including
the head, chest, arm, foot, and thigh (reviewed in Kavanagh and
tions is more consistent signals across individuals. Although such
accelerometers are inexpensive, they are often dedicated equip-
ment that has to remain attached at a particular location on the
Modern mobile phones have built-in accelerometers that can
be used to track movements without the need for an additional
device (Brezmes et al., 2009; Gyorbiro et al., 2009; Ryder et al.,
2009; Fernandes et al.,2011). They have their own power sources,
memory storage capabilities, and can transmit data wirelessly.
Patients could simply download an app onto their smartphone
enabling data collection and analysis. Mobile phones allow auto-
be invaluable to large-scale studies and personal patient health
Unfortunately, applying current mobile phone strategies to
populations with motor disabilities is challenging. For example,
muscles, loss of common automatic movements, and impaired
posture (Jankovic,2008). These symptoms all can adversely affect
activity recognition. Activity recognition strategies have been tai-
lored specifically for the elderly (Najafi et al., 2003), individuals
with muscular dystrophy (Jeannet et al., 2011), and even PD
November 2012 | Volume 3 | Article 158 | 1
Albert et al.Phone-based activity recognition for Parkinson’s
(Salarian et al., 2007), but each of these studies were done with
throughout the body. Performing activity recognition for a popu-
lation with a motor disability when carrying the phone naturally
in pockets or belt clips provides additional challenges.
Here we show how modern machine learning techniques can
quantify the movements of PD patients that carry mobile phones.
We first collect data from both healthy subjects and PD patients
performing a standard set of activities. From this data,we analyze
the precision of activity recognition both within and across the
two groups. Ultimately, we demonstrate an approach to activity
recognition using mobile phone devices for patient populations,
allowing us to better monitor patient responses to treatment in
MATERIALS AND METHODS
Eighteen healthy subjects and eight PD patients were recruited for
this study. The eighteen healthy subjects, having had no previous
subjects (6M/7F, 25.1±3.0years), and five older subjects (5F,
patients in Hoehn and Yahr stage 1–3 (7F/1M, 67.0±8.1years,
median±range). Patients were recorded while taking their
usual medications (ON-med condition). Many patients presented
mild dyskinesias during the course of the experiment. Written,
University Institutional Review Board approved this study.
All subjects were instructed to carry T-mobile G1 phones run-
ning Android OS version 1.6 in their front pockets. These phones
have a standard built-in tri-axial accelerometer with a range of
±2.8g. The sampling rate was variable between 15 and 25Hz
depending upon the amount of movement.
select the activity on a specially designed phone app (Figure 1)
with the experimenter present to minimize errors. The accelera-
tions were labeled according to the activity they were performing.
The activities were performed in a laboratory setting in the order
shown below. For a few initial subjects the activities began and
simplified by removing this activity for all subjects later on. We
also repeated activities to get more recordings with the phone in
slightly different positions on the subjects.
3. holding the phone (standing with arms bent forward)
5. not wearing (set on a table)
7. holding the phone
DATA PROCESSING AND CLASSIFICATION
The accelerometer signals were preprocessed using the follow-
ing procedure: the recordings were segmented into 10s clips
FIGURE 1 | Recording device and software. (A)The subjects carried
T-mobile G1 android phones in their pockets. (B)The axes of the
accelerometer relative to the orientation of the phone in (A). (C)The screen
which subjects selected which activity they were performing.
performing a given activity for the entire duration. The first and
time for the phones to enter and be removed from the pockets
of subjects. Three thousand three hundred and eighty eight sam-
were recorded from PD patients. The subject and researcher both
observed labeling during recording, ensuring validity of training,
The phone accelerometer values were linearly interpolated from a
variable rate between 15 and 25Hz to match 20Hz.
The subjects naturally placed the phone in their pockets in the
following possible orientations, due to the elongated rectangular
shape of the phones.
1. screen in/right side up
2. screen in/upside down
3. screen out/right side up
4. screen out/upside down
In the accelerometer readings, each of these orientations only
differs by the signs of the axes. For example, flipping the phone
right side up to upside down (e.g., orientation #1–#2) changes
the sign of the x and y axes, while turning the screen inward
(e.g., #1–#3) changes the sign of axes x and z axes. To cor-
rect for these different phone orientations, we generated three
form that effectively flips the phone 180˚ along each of its three
From these 10s clips features were extracted,as summarized in
Frontiers in Neurology | Movement Disorders
November 2012 | Volume 3 | Article 158 | 2
Albert et al. Phone-based activity recognition for Parkinson’s
Table 1 | Features used for activity recognition.
Mean, absolute value of the mean
Moments: standard deviation, skew, kurtosis
For the change in acceleration: mean, standard
deviation, skew, kurtosis
Root mean square
Smoothed root mean square (5pt kernel, 10pt kernel)
Extremes: min, max, abs min, abs max
Histogram: includes counts for −4 to 4 z-score bins
Fourier components: 32 samples for each axis
Overall mean acceleration
Cross product means: xy, xz, yz
Abs mean of the cross products
Two popular algorithms were used for classification: support
ized) multinomial logistic regression sparse multinomial logistic
regression (SMLR; Krishnapuram et al., 2005). Both techniques
large feature sets. The hyperparameters for both classifiers were
found by a grid search of 10× where x is an integer between −5
ing the healthy subject labeled activities. Given the size of the data
set used for cross validation this procedure was not expected not
the coefficient for the regularization term during optimization,λ,
was 0.0001. For SVM,we normalized each feature to have 0 mean
and unit variance. We applied radial basis functions, giving us
two hyperparameters – the soft slack variable, C, and the size of
the Gaussian kernel, γ. The values found by cross validation were
C =1 and γ=0.1 for the across subjects validation and C =10
and γ=1 for the 10-fold validation.
To examine our ability to classify activities in a patient popula-
tion, we collected data on both PD patients and healthy subjects,
as detailed in the Section“Materials and Methods.”Subjects were
performing a series of activities (Figure 1C). We applied two dif-
ferent classifiers, SVM and SMLR, to classify the activities. Our
intention is to demonstrate the importance of using classifiers
trained with data specifically from patient populations.
the accelerometer signals. Recordings were made from the three-
axis accelerometers in the phones. The orientation of the phone
determined the orientation of the accelerometer axes (Figure1B).
There are also visible differences in the data of PD patients com-
pared to healthy subjects. Example clips from these activities
(Figure 2) show the presence of dyskinesias in one PD patient.
It can also be observed in the accelerations that walking is often
less periodic for PD patients than healthy subjects. Such dif-
ferences can lead to errors in classification if features such as
Table 2 | Classification matrix for healthy subjects with 10-fold cross
Activity Walking StandingHolding SittingNot wearing
96.1% overall accuracy for 18 healthy subjects.
periodicity or vibrations are used for prediction. Although the
movements are related between groups, the exact characteristics
of those movements have enough differences between the popu-
lations to examine the effect of these differences on classification
Unlike many studies which classify signals based on only a few,
specific features, we did not seek individual features that could
be used for classification. For example, knowing if someone is
standing, sitting, or holding the phone can depend not only on
the orientation of the phone, but also on the amount of vibra-
tion in the movement. Instead of searching for particular features
with clear,independent differences between activities,we chose to
apply the standard, state-of-the-art machine learning approach:
we constructed a large feature set and had the algorithms select
how to combine and weigh the features appropriately.
First, we wanted to compute a classification accuracy measure
that can be compared across studies. To do this, we used 10-fold
cross validation, selecting every 10th sample for the test set. This
accuracy is expected to be fairly high considering movement pat-
terns specific to individual subjects were in both the training and
test sets. For SVM classification, this lead to a 96.1% accuracy for
healthy subjects (Table 2), and a 92.2% accuracy for PD patients
and 84.7% for PD patients) so only the SVM results are shown for
We sought to quantify the effect of differences between move-
ments made by Parkinson’s patients and healthy subject on the
classification algorithm. Unlike previous studies that have ana-
lyzed the difference from healthy subjects for particular features
(Salarian et al., 2007; Jeannet et al., 2011), we directly trained the
classifiers using healthy subject data, and applied the classifiers to
patient recordings to observe the effect of those differences. From
this, we achieved a much lower accuracy of 60.3% using SVM’s
of the difference between the populations and the need to find
another, more accurate way to identify patient activities.
One entry in the table is made for every 10s sample. The rows
represent the activity being performed and the columns represent
the prediction according to the algorithm. Overall accuracy is the
sum of the correct classification (diagonal in bold) compared to
the sum of all entries in the entire table.
for the within vs. across population results. Ten-fold or leave-one-
out cross validation techniques do not remove the effect of the
same individual that may have significantly distinct movement
November 2012 | Volume 3 | Article 158 | 3
Albert et al.Phone-based activity recognition for Parkinson’s
HoldingSitting Standing Walking
FIGURE 2 |Typical examples of accelerometer readings for Parkinson’s
patients and healthy subjects for the four activities studied. Red, green,
and blue lines are the x, y, and z-axis accelerations, as specified in Figure 1B.
The patient shown here exhibited dyskinesia in the arm that is clearly visible
while holding the phone and somewhat visible during standing and sitting.
The patient also had an irregular gait cycle during walking.
Table 3 | Classification matrix for PD patients with 10-fold cross
ActivityWalkingStanding Holding SittingNot wearing
92.2% overall accuracy for eight PD subjects.
Table 4 | Classification matrix for PD patients using healthy subject
ActivityWalking StandingHoldingSittingNot wearing
60.3% overall accuracy for 18 healthy subjects for training and eight PD patients
pattern from others. Using 10-fold cross validation would still
allow such idiosyncratic movements of individuals to be part of
the accuracy of the algorithm if it were applied to subjects after
training. For this we performed subject-wise cross validation for
the 18 healthy subjects and found an accuracy of 86.0% for SVM’s
and 85.2% for SMLR. Note that because much of the variation
in movements is across subjects,this accuracy is much lower than
that of the 10-fold cross validation. Table 5 presents a breakdown
of the classification for SVM’s, with SMLR appearing similar.
To consider the ability of this approach to be adapted to PD
patients by using patient data, we also analyzed the PD patients
Table 5 | Classification matrix for healthy subjects with subject-wise
ActivityWalkingStanding HoldingSitting Not wearing
86.0% overall accuracy for 18 healthy subjects.
Table 6 | Classification matrix for PD patients with subject-wise cross
75.1% overall accuracy for eight PD subjects.
separately. Similar to the healthy subjects,we applied subject-wise
cross validation on the PD patient data alone. Using SVM’s, the
accuracy was 75.1% (Table 6) and using SMLR it was 76.0%.
Although this is lower than the previous percentage for healthy
subjects, this is expected as PD patients movements vary more
significantly across subjects. Most importantly, when consider-
ing predictions across subjects, training using patient data led to
a significantly better prediction than training using healthy data
We applied machine learning to signals from mobile phones to
Frontiers in Neurology | Movement Disorders
November 2012 | Volume 3 | Article 158 | 4
Albert et al.Phone-based activity recognition for Parkinson’s
set and had the relevant features selected by the machine learning
these methods were not expected to test well for populations with
unique movement patterns. This was done using mobile phones
rying was expected to lower accuracy values, it is more indicative
of expectedaccuracyif thisresearchistobeappliedtostudieswith
as they are small, relatively inexpensive, and register both move-
ment and orientation to gravity. Some systems have integrated
temperature, compass, light, and sound sensors on the waist
(Choudhury et al., 2008) or a similar collection of multimodal
sensors on the wrist (Maurer et al., 2006; Gyorbiro et al., 2009).
For accelerometer-only arrays, multiple sensors may be placed
throughout the body – anywhere from three to five locations (Bao
and Intille, 2004; Tapia et al., 2007; Krishnan and Panchanathan,
2008) or more. Mannini and Sabatini (2010) provide a review of
There are simpler alternatives to using multiple sensors,
improving the convenience, cost, and compliance rates. The most
common approach is to use a single, waist-mounted accelerom-
eter. This approach has been analyzed on very specific sets of
instructed activities with over 98% accuracy (Mathie et al.,2004a;
Mathie et al.,2004b; Ravi et al.,2005; Lee et al.,2009). High accu-
racy ratings were possible in part due to the fixed location of the
accelerometers on the body, the use of within-subject vs. across-
subject cross validation, and the artificial nature of instructed
movements. Signals from walking,sitting,and standing are neces-
sarily more repeatable when in a consistent lab setting following
instruction. For comparison, when subjects simply wore such a
device for 24h, with more natural activities, accuracy was closer
been well-studied in the domain of activity recognition, but may
need consistent placement for high accuracy.
Unlike dedicated accelerometers, some people already con-
sistently carry mobile phones, making them a convenient plat-
form for recording movements. Most smartphones have built-in
accelerometers and are often worn on the person, similar in prin-
ciple to previous work on accelerometry. Mobile phones have
built-in communication protocols that allow simple data logging
to the device and wireless transmission. This permits real-time
response, or in an experimental setting, compliance verification.
Because mobile phones are widely adopted, compliance without
verification is already high, as people are used to carrying them.
Due to these advantages, mobile phones have the promise to pro-
vide a convenient, inexpensive, and objective means to detect the
activities of people.
Mobile phones have been used to classify activities of healthy
subjects (Bieber et al., 2009; Brezmes et al., 2009; Gyorbiro et al.,
2009; Ryder et al., 2009; Wang et al., 2009; Yang, 2009; Kwapisz
standing, sitting, and using stairs. The choice of activities influ-
studies are not subject-wise cross-validated, applicability across
subjects is more difficult to interpret. In Kwapisz et al. (2011),
healthy subjects were instructed to carry the phone in their left
pocket and perform a specific set of activities; all activities except
stair climbing were classified with at least 90% accuracy. Other
studies found similarly high accuracy but with different classifi-
cation techniques (Brezmes et al., 2009; Ryder et al., 2009; Yang,
2009). In Wang et al. (2009), classes were divided as still, walking,
running,or in a vehicle,which simplified classification which was
done using microphones and GPS as well as accelerometer read-
ings. In Yang (2009), a preprocessing technique was used which
converted the axes from phone-specific to phone-independent
coordinates based on orientation of gravity, providing 88–90%
accuracy. While our results on healthy subjects are in line with
previous studies, the central contribution of our paper is the
careful analysis of precision of activity recognition in the context
We chose to analyze the PD population for various reasons. Mil-
lions of people throughout the world are suffering from diseases
that affect mobility. Many diseases, such as stroke, heart disease,
or depression affect large populations but have a wide variety of
causes, types, and symptoms. PD, on the other hand, is char-
acterized by a number of common characteristics, which makes
analysis easier across subjects (Gelb et al.,1999). Common symp-
well to analyses using accelerometers in mobile phones (Joundi
et al., 2011; Surangsrirat and Thanawattano, 2012). The PD pop-
ulation is also an important subgroup to consider as it also effects
a relatively large population – approximately four million people
globally (Dorsey et al., 2007).
There is another study that automatically classified and char-
acterized postures and activities for a population of PD patients
(Salarian et al., 2007). However, their results used within-subject
cross validation and thus cannot speak to the across-subject gen-
eralization issue we are discussing here. Moreover, they used a
set of accelerometers and gyroscopes instead of mobile phones.
Our paper demonstrates the ability to use mobile phone record-
ings of acceleration to enable quality activity recognition with PD
There are a few limitations to the interpretation of our results
to address. For our healthy subjects,we used a population of both
of the difference between the groups can be age-related, however
we believe this effect was minor compared to the effect of PD on
patient movements.Also,the PD group was relatively small (eight
ment by using PD training data. Lastly, because we had both the
the instructed actions in a typical fashion (e.g., moving feet while
possible inconsistencies by hand,and thus affecting the validity of
limitations, the main conclusions of this study are supported.
November 2012 | Volume 3 | Article 158 | 5
Albert et al.Phone-based activity recognition for Parkinson’s
There were two main goals for this study. First, we demon-
strate how machine learning can be used to infer the activities
of PD populations; the focus is not on particular, hand-picked
features of movement, but on automated methods of weighing
and combining those features. The second major goal was to
highlight and quantify the effect of applying classifiers designed
for healthy subjects on a PD patient population. A demonstra-
ble drop in classification accuracy from 92.2 to 60.3% makes this
point clear; it is important to use tools and analyses designed for
specific patient populations. Although this study is not thorough
enough to validate this classification method for clinical practice,
it does demonstrate a strong benefit of machine learning, and a
methods designed for healthy subjects.
The ultimate objective of therapies is to improve patient qual-
ity of life and activity tracking is an additional way of quantifying
test and optimize aspects of many therapies for motor disorders.
By only downloading an application, mobile phones can record
a person’s movements, greatly simplifying the study design and
improving compliance. This information can be of personal or
community medical use, improving evaluation of patient out-
motor impairments require special consideration in approaches
that analyze movement patterns. Mobile phones provide a means
of tracking movements in an objective, convenient, and inex-
pensive way. The extent to which leveraging these qualities can
improve and enable new therapeutic approaches is an area of
This work was supported in part by the National Parkinson Foun-
dation and the U.S. National Institutes of Health under Grants
Andrew Cichowski and Aaron Yang for their help in recruiting
subjects and data collection.
Bao, L., and Intille, S. (2004). “Activ-
ity recognition from user-annotated
puting, eds A. Ferscha and F. Mat-
tern (Berlin/Heidelberg: Springer),
Bieber, G., Voskamp, J. R., and Urban,
B. (2009). “Activity recognition for
versal access,” in Human-Computer
Interaction. Intelligent and Ubiqui-
tous Interaction Environments, ed.
C. Stephanidis (Berlin/Heidelberg:
Brezmes, T., Gorricho, J.-L., and Cot-
rina, J. (2009). “Activity recogni-
tion from accelerometer data on
a mobile phone,” in IWANN ‘09
Proceedings of the 10th Interna-
tional Work-Conference on Artifi-
cial Neural Networks: Part II: Dis-
tributed Computing, Artificial Intel-
ligence, Bioinformatics, Soft Comput-
ing, and Ambient Assisted Living, eds
S. Omatu, M. Rocha, J. Bravo, F.
Fernández, E. Corchado, A. Bustillo
and J. Corchado (Berlin/Heidelberg:
SVM: a library for support vector
machines. ACM Trans. Intell. Syst.
Technol. 2, 1–27.
Choudhury, T., Consolvo, S., Harri-
son, B., Hightower, J., Lamarca,
A., Legrand, L., et al. (2008). The
mobile sensing platform: an embed-
ded activity recognition system. Per-
vasive Comput. IEEE 7, 32–41.
Dibble, L. E., Addison, O., and Papa, E.
ease: a systematic review across the
disability spectrum. J. Neurol. Phys.
Ther. 33, 14–26.
Dorsey, E. R., Constantinescu, R.,
Thompson, J. P., Biglan, K. M.,
et al. (2007). Projected number
of people with Parkinson disease
in the most populous nations,
2005 through 2030. Neurology 68,
Ellis, T., Cavanaugh, J. T., Earhart,
G. M., Ford, M. P., Foreman,
K. B., and Dibble, L. E. (2011).
and motor impairment best predict
quality of life in Parkinson‚Äôs dis-
ease? Parkinsonism Relat. Disord. 17,
Fernandes, H.L., Albert, M. V.,
phones. PLoS ONE
Gelb, D. J., Oliver, E., and Gilman,
S. (1999). Diagnostic criteria for
Parkinson disease. Arch. Neurol. 56,
Goodwin, V. A., Richards, S. H., Tay-
lor, R. S., Taylor, A. H., and Camp-
bell, J. L. (2008). The effectiveness
of exercise interventions for people
with Parkinson’s disease: a system-
atic review and meta-analysis. Mov.
Disord. 23, 631–640.
Homanyi, G. (2009). An activ-
ity recognition system for mobile
phones. Mobile Netw. Appl. 14,
Jankovic, J. (2008). Parkinson’s dis-
ease: clinical features and diagnosis.
J. Neurol. Neurosurg. Psychiatr. 79,
Jeannet, P.-Y., Aminian, K., Bloetzer,
G., Kieburtz, K.,
A. (2011). Continuous monitoring
and quantification of multiple para-
meters of daily physical activity
in ambulatory Duchenne muscular
dystrophy patients. Eur. J. Paediatr.
Neurol. 15, 40–47.
Joundi, R. A., Brittain, J.-S., Jenkinson,
N., Green,A. L., and Aziz, T. (2011).
Rapid tremor frequency assess-
ment with the iPhone accelerome-
ter. Parkinsonism Relat. Disord. 17,
Kavanagh, J. J., and Menz, H. B.
(2008). Accelerometry: a technique
for quantifying movement patterns
during walking. Gait Posture 28,
Krishnan, N. C., and Panchanathan,
S. (2008). “Analysis of low resolu-
tion accelerometer data for contin-
uous human activity recognition,”
in International 28 Conference on,
Acoustics,Speech,and Signal Process-
ing, ICASSP 2008 IEEE, Las Vegas,
M. A. T., and Hartemink, A. J.
(2005). Sparse multinomial logis-
tic regression: fast algorithms and
generalization bounds. IEEE Trans.
Kwapisz, J. R., Weiss, G. M., and
accelerometers. SIGKDD Explor. 12,
accelerometer,” in Proceedings of the
World Congress on Engineering and
Computer Science, San Francisco.
Long, X., Yin, B., and Aarts, R. M.
daily physical activity classification,”
in Annual International Conference
of the IEEE Engineering in Medi-
cine and Biology Society EMBC 2009,
Machine learning methods for clas-
on-body accelerometers. Sensors 10,
Mathie, M. J., Celler, B. G., Lovell,
N. H., and Coster, A. C. (2004a).
Classification of basic daily move-
ments using a triaxial accelerom-
eter. Med. Biol. Eng. Comput. 42,
Mathie, M. J., Coster, A. C. F., Lovell,
N. H., and Celler, B. G. (2004b).
Accelerometry: providing an inte-
grated, practical method for long-
term, ambulatory monitoring of
human movement. Physiol. Meas.
Maurer, U., Smailagic, A., Siewiorek, D.
P., and Deisher, M. (2006). “Activity
recognition and monitoring (using)
multiple sensors on different body
on Wearable and Implantable Body
Sensor Networks, BSN 2006,Aachen,
Najafi, B., Aminian, K., Paraschiv-
Ionescu, A., Loew, F., Bula, C. J.,
and Robert, P. (2003). Ambulatory
system for human motion analysis
using a kinematic sensor: monitor-
ing of daily physical activity in the
elderly. IEEETrans. Biomed. Eng. 50,
Palmer, S. S., Mortimer, J. A., Web-
ster, D. D., Bistevins, R., and
Dickinson, G. L. (1986). Exer-
cise therapy for Parkinson’s dis-
ease. Arch. Phys. Med. Rehabil. 67,
Frontiers in Neurology | Movement Disorders
November 2012 | Volume 3 | Article 158 | 6
Albert et al.Phone-based activity recognition for Parkinson’s Download full-text
Ravi, N., Dandekar, N., Mysore, P., and
Littman, M. (2005).“Activity recog-
nition from accelerometer data,” in
Proceedings of the Seventeenth Con-
ference on Innovative Applications
of Artificial Intelligence, (Pittsburgh:
AAAI Press), 1541–1546.
Ryder, J., Longstaff, B., Reddy, S.,
and Estrin, D. (2009). “Ambula-
tion: a tool for monitoring mobil-
ity patterns over time using mobile
phones,”in International Conference
on Computational Science and Engi-
F. J. G., Burkhard, P. R., and Amin-
ing of physical activities in patients
Biomed. Eng. 54, 2296–2299.
hatla, M., Chandler, J., Pieper, C.
F., Ray, L., et al. (1998). Exercise
to improve spinal flexibility and
function for people with Parkin-
son’s disease: a randomized, con-
trolled trial. J. Am. Geriatr. Soc. 46,
Stone, A. A., Shiffman, S., Schwartz, J.
E., Broderick, J. E., and Hufford, M.
R. (2003). Patient compliance with
Clin. Trials 24, 182–199.
Surangsrirat, D., and Thanawattano, C.
ral analysis in Parkinson’s Disease,”
in Proceedings of IEEE Southeastcon,
Tapia, E. M., Intille, S. S., Haskell, W.,
Larson, K., Wright, J., King, A., et
al. (2007).“Real-time recognition of
physical activities and their inten-
sities using wireless accelerometers
and a heart rate monitor,” in Pro-
ceedings of the 2007 11th IEEE Inter-
national Symposium on Wearable
Computers. IEEE Computer Society,
B., et al. (2009). “A framework of
energy efficient mobile sensing for
automatic user state recognition,”in
Proceedings of the 7th international
conference on Mobile systems, appli-
Yang, J. (2009). “Toward physical activ-
ity diary: motion recognition using
simple acceleration features with
mobile phones,” in Proceedings of
the 1st international workshop on
Interactive multimedia for consumer
electronics, Beijing, China: ACM.
Conflict of Interest Statement: The
authors declare that the research was
conducted in the absence of any com-
mercial or financial relationships that
could be construed as a potential con-
flict of interest.
Received: 18 April 2012; accepted: 19
October 2012; published online: 07
Citation: Albert MV, Toledo S, Shapiro
M and Kording K (2012) Using mobile
phones for activity recognition in Parkin-
son’s patients. Front. Neur. 3:158. doi:
This article was submitted to Frontiers
in Movement Disorders, a specialty of
Frontiers in Neurology.
and Kording. This is an open-access arti-
cle distributed under the terms of the
Creative Commons Attribution License,
which permits use, distribution and
reproduction in other forums, provided
the original authors and source are cred-
ited and subject to any copyright notices
concerning any third-party graphics etc.
November 2012 | Volume 3 | Article 158 | 7