Privacy Implications of Accelerometer Data: A Review of Possible Inferences

Abstract

Accelerometers are sensors for measuring acceleration forces. They can be found embedded in many types of mobile devices, including tablet PCs, smartphones, and smartwatches. Some common uses of built-in accelerometers are automatic image stabilization, device orientation detection, and shake detection. In contrast to sensors like microphones and cameras, accelerometers are widely regarded as not privacy-intrusive. This sentiment is reflected in protection policies of current mobile operating systems, where third-party apps can access accelerometer data without requiring security permission. It has been shown in experiments, however, that seemingly innocuous sensors can be used as a side channel to infer highly sensitive information about people in their vicinity. Drawing from existing literature, we found that accelerometer data alone may be sufficient to obtain information about a device holder's location, activities, health condition, body features, gender, age, personality traits, and emotional state. Acceleration signals can even be used to uniquely identify a person based on biometric movement patterns and to reconstruct sequences of text entered into a device, including passwords. In the light of these possible inferences, we suggest that accelerometers should urgently be re-evaluated in terms of their privacy implications, along with corresponding adjustments to sensor protection mechanisms.

Full text

Privacy Implications of Accelerometer Data:
A Review of Possible Inferences
Jacob Leon Kröger
Technische Universität Berlin
Hardenbergstraße 32
10623 Berlin
+49 30 7001 41037
kroeger@tu-berlin.de
Philip Raschke
Technische Universität Berlin
Ernst-Reuter-Platz 7
10587 Berlin
+49 30 8353 58353
philip.raschke@tu-berlin.de
Towhidur Rahman Bhuiyan
Technische Universität Berlin
Hardenbergstraße 32
10623 Berlin
+49 30 7001 41001
t.bhuiyan@campus.tu-berlin.de
ABSTRACT
Accelerometers are sensors for measuring acceleration forces.
They can be found embedded in many types of mobile devices,
including tablet PCs, smartphones, and smartwatches. Some
common uses of built-in accelerometers are automatic image
stabilization, device orientation detection, and shake detection. In
contrast to sensors like microphones and cameras, accelerometers
are widely regarded as not privacy-intrusive. This sentiment is
reflected in protection policies of current mobile operating
systems, where third-party apps can access accelerometer data
without requiring security permission. It has been shown in
experiments, however, that seemingly innocuous sensors can be
used as a side channel to infer highly sensitive information about
people in their vicinity. Drawing from existing literature, we
found that accelerometer data alone may be sufficient to obtain
information about a device holder’s location, activities, health
condition, body features, gender, age, personality traits, and
emotional state. Acceleration signals can even be used to uniquely
identify a person based on biometric movement patterns and to
reconstruct sequences of text entered into a device, including
passwords. In the light of these possible inferences, we suggest
that accelerometers should urgently be re-evaluated in terms of
their privacy implications, along with corresponding adjustments
to sensor protection mechanisms.
CCS Concepts
• Security and privacy
Keywords
Accelerometer, Sensor, Privacy, Side channel, Inference attack
1. INTRODUCTION
An accelerometer is an instrument for measuring acceleration
forces caused by the movements and vibrations of an object, or by
gravity. Today, all sorts of mobile devices, including smart-
phones, tablet PCs, smartwatches, digital cameras, wearable
fitness trackers, game controllers, and virtual reality headsets, are
equipped with built-in microelectromechanical accelerometers [1].
Studies even suggest that accelerometers are the most widely used
sensor in wearable devices [2] and also the sensor that is most
frequently accessed by mobile apps [3].
Among other common applications, acceleration signals are used
for image stabilization in cameras, for measuring the orientation
of a device relative to Earth’s gravitational pull (e.g. to enable
automatic display rotation between landscape and portrait mode),
and for detecting user actions, such as moving or shaking a
device.
While some sensors, such as microphones, cameras and GPS, are
widely perceived as privacy-sensitive [4, 5] and require explicit
user permission to be activated in current mobile operating
systems [3], accelerometers are less well-understood in terms of
their privacy implications, and also much less protected [6, 7].
Even scholarly literature has largely ignored potential issues in
this field, with researchers describing accelerometer data as “not
particularly sensitive” [8] or even “privacy preserving” [9].
Experimental studies have shown, however, that sensitive
personal data can be inferred from accelerometer readings. This
paper presents a non-exhaustive overview of possible inferences,
drawing from multiple academic disciplines, including infor-
mation science, psychology, health science, and computer science.
According to our findings, accelerometers in mobile devices may
reveal information about a user’s activities (section 2.1), location
(sect. 2.2), identity (sect. 2.3), device inputs (sect. 2.4), health
condition and body features (sect. 2.5), age and gender (sect. 2.6),
moods and emotions (sect. 2.7), and personality traits (sect. 2.8).
2. POSSIBLE INFERENCES
In this chapter, we present experimental studies from the scholarly
literature in which sensitive information was successfully derived
from accelerometer data. A visual overview is provided in Fig. 3,
at the end of the chapter.
2.1 Activity and Behavior Tracking
A wide range of physical activity variables and behavior-related
information can be derived from raw accelerometer data.
Accelerometer-based pedometers (“step counters”), for instance,
register the impacts produced by steps during motion and can
estimate energy expenditure and distance walked [10]. In medical
studies, wearable devices with embedded accelerometers are
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. Copyrights
for components of this work owned by others than the author(s) must be
honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee. Request permissions from
Permissions@acm.org.
ICCSP 2019, January 19–21, 2019, Kuala Lumpur, Malaysia.
© 2019 Copyright is held by the owner/author(s). Publication rights
licensed to ACM.
ACM ISBN 978-1-4503-6618-2/19/01…$15.00
DOI: https://doi.org/10.1145/3309074.3309076
81

widely used to assess the amount of sedentary time and physical
activity among patients [11, 12].
Body-worn accelerometers have also been shown to enable real-
time body posture and activity classification. High recognition
accuracy has been achieved for basic physical activities, including
running, walking, cycling, lying, climbing stairs, falling, sitting
and standing [13–16], as well as for more complex activities, such
as writing, reading, typing, painting, sorting paperwork or
searching the internet [17].
Not only the type but also the duration of activities and temporal
behavior patterns can be derived from acceleration signals [18,
19]. When worn during the night, mobile devices with built-in
accelerometers may enable sleep-wake cycle monitoring, through
variables such as sleep onset and offset, total sleep time and sleep
intervals [20, 21], as well as the monitoring of sleep-related
behaviors [11].
Accelerometers in handheld and wrist-worn devices can further be
used to detect specific hand gestures [22], eating and drinking
moments [23, 24], and smoking [25, 26]. Gait features of subjects,
extracted from accelerometer data, can even reveal their level of
intoxication. Researchers were able to distinguish “sober walk”
from “intoxicated walk” [27] and to estimate blood alcohol
content [28] as well as the number of drinks consumed [29] via
accelerometry alone.
In [17], signals from a single body-worn accelerometer were used
to detect if a subject is carrying a load. Accelerometer-based gait
dynamics have also been used to estimate the weight of carried
objects with robustness to variations in walking speeds, body
types and walking conditions [30].
Figure 1: Classification of driving patterns based on streams
of accelerometer data, from [31].
When located inside a car, motion sensors can be used to measure
an operator’s driving behavior. In [31], Singh, Juneja and Kapoor
identified events such as sudden breaking, sudden acceleration,
right and left turns and lane changes from patterns in
accelerometer data, as is illustrated in Fig. 1. From such infor-
mation, researchers were able to detect aggressive or unsafe
driving styles [32] and drunk driving patterns [33].
Based on indicative body movements and sound vibrations, both
measured using accelerometers, researchers were able to derive
speech activity and social interactions of subjects [9, 34]. Even
ways of reconstructing speech solely from recorded vibrations
have been explored. AccelWord, developed in [35], can detect
hotwords spoken by a user, utilizing accelerometer data from
commercially available mobile devices. Patents have already been
filed for a “method of detecting a user's voice activity using an
accelerometer” [36] and a “system that uses an accelerometer in a
mobile device to detect hotwords” [37].
2.2 Location Tracking
It has been shown that accelerometers in mobile devices can be
exploited for user localization and reconstruction of travel
trajectories, even when other localization systems, such as GPS,
are disabled. In [38], Han et al. were able to geographically track
a person who is driving a car based solely on accelerometer
readings from the subject’s smartphone. In their approach, they
first calculate the vehicle’s approximate motion trajectory using
three-axis acceleration measurements from an iPhone located
inside the vehicle, and then map the derived trajectory to the
shape of existing routes on a map. An example application of the
algorithm is displayed in Fig. 2. Han et al. describe their results as
“comparable to the typical accuracy for handheld global
positioning systems.”
Figure 2: Map matching algorithm used in [38]. The green
trail indicates the motion trajectory obtained from
accelerometer data. The red trail indicates the inferred route.
The blue trail indicates the actual route traveled (GPS data).
Hua, Shen and Zhong found that accelerometers in smartphones
can also reveal the device’s location while the holder is using a
metropolitan train system [39]. To achieve this, the researchers
compare and match acceleration patterns with labeled training
data to recognize specific station intervals through which the user
travels. Results from experiments on a real metro line show that
the accuracy of their approach could reach up to 89% and 92% if
the metro ride is longer than 3 or 5 stations, respectively [39].
2.3 User Identification
Body movement patterns recorded by accelerometers in mobile
devices have been demonstrated to be discriminative enough to
differentiate between, or even uniquely identify, users. Various
accelerometer-only approaches have been proposed to confirm the
identity of a user based on biometric gait features [40, 41], hand
gestures [42], or head movements [43]. Using accelerometer rea-
82

dings from smartphones, Kwapisz, Weiss and Moore were able to
recognize individuals from a pool of 36 test subjects with 100%
accuracy [44].
It has also been shown that, through aerial vibrations, accelero-
meters can be sensitive enough to capture sound, including human
speech, in sufficient quality to distinguish between different spea-
kers with high accuracy [35].
The location trajectory of a mobile device, which can be inferred
from accelerometer data under certain conditions (as explained in
section 2.2), may reveal a user’s work and home addresses [45],
and – in conjunction with white pages, employment directories,
tax records, or other auxiliary datasets – a user’s real identity [46].
Following an approach commonly referred to as device
fingerprinting, users can further be told apart based on unique
characteristics and features of their personal devices. Calibration
errors in accelerometers, which are caused by imperfections in the
manufacturing process, have been found sufficient to uniquely
identify their encapsulating device [6, 47]. Such a “fingerprint”
can be used, for instance, to track users across repeated website
visits, even when private browsing is activated and other tracking
technologies, such as canvas fingerprinting or cookies, are
blocked [48].
2.4 Keystroke Logging
The input that users type into to their devices through
touchscreens and keyboards contains highly sensitive information
such as text messages, personal notes, login credentials and
transaction details.
Based on the observation that swipes, taps and keystrokes often
correlate with distinctive hand movements of the user, it has been
shown that inputs can be reconstructed using motion sensor data
from handheld and wrist-worn devices [49–51]. Some researchers
have exclusively used accelerometer data for such keystroke
inference attacks. Aviv et al. demonstrated that accelerometers in
smartphones can be exploited to infer tap- and gesture-based
input, including PINs and graphical password patterns [52]. Based
on the same type of data, Owusu et al. were able to obtain entire
sequences of text entered through a phone’s touchscreen [53].
Through examining the source code of other existing approaches,
it has been found that even multi-sensor attacks solely use
acceleration information for tap detection, leading to the
conclusion that defense mechanisms against these kinds of side
channel attacks should focus on accelerometers [54].
Not only does the above imply that accelerometer data could offer
sensitive insights into a user’s communication and transactions:
Beltramelli and Risi even warn that a user’s entire technological
ecosystem could be compromised when passwords are leaked
through embedded sensors in consumer electronics [55].
2.5 Inference of Health Parameters and Body
Features
Body-worn accelerometers can be used to gain insight into a
person’s physical characteristics and health status. Using
accelerometer data from smartphones, researchers were able to
derive an approximation of the body weight and height of users
[56, 57]. A strong correlation has been observed between
accelerometer-determined physical activity and obesity [58].
Physical activity is generally recognized as a promoter and
indicator of health [59]. A person’s amount of physical activity
can reveal sensitive information about latent chronic diseases and
the person’s degree of mobility [12] as well as about cognitive
function and even risk of mortality [60]. As explained in section
2.1, a wide range of activity-related variables can be derived from
accelerometer data, including energy expenditure, type of activity
and temporal activity patterns. This association is increasingly put
to use in health studies, where accelerometers are used to
remotely assess the physical activity level of participants [61].
Another important factor in population health is the amount of
sleep that people get. Sleep loss has been associated with
developing serious illnesses, such as cardiovascular disease and
diabetes, and even with increased all-cause mortality [62].
Numerous studies have shown that accelerometers in wearable
devices can be used for evaluating sleep patterns [20], sleep
fragmentation [63] and sleep efficiency [64]. Actigraphy, an
accelerometer-based assessment method, has been described as an
“essential tool in sleep research and sleep medicine” [20].
Experimental results from Pesonen and Kuula suggest that
accelerometers in consumer-targeted wearables can be as effective
for sleep monitoring as research-targeted devices [21].
Specialized accelerometers have been used to measure various
other health parameters, including voice health [65], postural
stability [12] and physiological sound [66].
2.6 Inference of Demographics
Estimates of demographic variables such as age and gender can be
made based on data from body-worn accelerometers. It has long
been demonstrated that adults and children differ in their
smoothness of walking, which is reflected in accelerometer
readings [67]. Menz, Lord and Fitzpatrick compared gait features
between young and elder subjects using acceleration signals and
discovered that younger subjects showed greater step length,
higher velocity and smaller step timing variability [68]. Using
data from accelerometers in smartphones, Davarci et al. were able
to predict the age interval of test subjects with a success rate of
92.5% [69]. Their work is based on the observation that children
and adults differ in the way they hold and touch smartphones.
Experimental results by Cho, Park and Kwon indicate that there
are also gender-specific movement patterns [70]. In accordance,
research has shown that it is possible to estimate the sex of
individuals based on hip movements [56], gait features [71] and
physical activity patterns [72], all derived from accelerometer
data. An experiment also revealed that female gait patterns are
significantly influenced by the heel height of their shoes [73].
Weiss and Lockhart emphasize that accelerometer-based gender
recognition can work independently of a subject’s weight and
height [56]. Even acoustic vibrations caused by a person’s voice
and captured through a smartphone accelerometer can be used to
classify speakers into male and female with high accuracy [35].
2.7 Mood and Emotion Recognition
The level of physical activity, which can be measured using body-
worn accelerometers (see section 2.1), has been identified as a
potential predictor of human emotions [74] and depressive
moods [75]. Zhang et al. were able to recognize emotional states
of test subjects (happy, neutral, and angry) with fair accuracy,
relying only on accelerometer data from smart wristbands [76].
Accelerometers in smartphones have been used to detect stress
levels [77] and arousal [78] in users. Also, Matic et al. found a
positive association between accelerometer-derived speech activi-
ty and mood changes [9].
2.8 Inference of Personality Traits
Methods have been proposed for inferring preferences and other
personality traits solely from body gestures and motion patterns.
Englebienne and Hung used wearable accelerometers to estimate
the motivations, interests and group affiliations of study
83

participants in scenarios of social interaction, based on their
movements, body postures and expansiveness of gesturing [34].
A person’s level of physical activity, which can also be measured
using body-worn accelerometers (see section 2.1), has been shown
to correlate with certain personality traits such as conscientious-
ness, neuroticism, openness, and extraversion [79]. Artese et al.
evaluated the body movements of test subjects for seven days
using accelerometer-based monitoring devices and found that
agreeableness, conscientiousness and extraversion were positively
and neuroticism negatively associated to more steps per day and
other physical activity variables [80]. Examining correlates
between the personality and physical activity of female college
students, Wilson et al. discovered that neuroticism and
the functioning of the behavioral inhibition system were both
related to physical activity measures derived from accelerometer
readings [81].
3. DISCUSSION AND IMPLICATIONS
As shown in the previous section, accelerometers in mobile
devices can allow serious invasions of user privacy. Even when
other sensors, such as cameras, microphones and GPS are turned
off, accelerometer data can be sufficient to obtain information
about a device holder’s location, health condition, body features,
age, gender, emotions and personality traits. Acceleration signals
may even be used to uniquely identify a person based on
biometric movement patterns and to reconstruct sequences of text
entered into a device.
It has to be acknowledged that most experimental studies cited in
this paper have substantial limitations. First, many approaches
were only tested in controlled laboratory settings [14, 17, 24, 26,
32, 33, 35, 40, 41, 43, 53, 57]. For methods applied under real-life
conditions, considerable reductions in accuracy have been
observed [9, 82]. Second, several of the presented methods require
prior knowledge about the user or the user’s context in order to
function [39–44, 52]. Third, subjects in some of the experiments
wore accelerometers attached to certain body parts, such as chest
[9, 15], hip [40], waist [14], or even multiple body parts [24, 25,
64], whereas in reality, mobile devices are mostly worn around
the wrist [23] or interchangeably in hands, bags, and pockets [83].
In light of these limitations, the real-world applicability of the
presented methods can be questioned.
On the other hand, it may reasonably be assumed that at least
some of the parties who regularly access accelerometer data from
consumer devices (e.g. device manufacturers, service providers,
app developers) possess larger sets of training data, more
technical expertise and more financial resources than the
researchers cited in this paper. Furthermore, data from other
sensors and auxiliary data may be available to potential
adversaries, improving their capability to draw sensitive
inferences, while the methods considered in this paper solely rely
on accelerometer data. Thus, our work represents only an initial
and non-exhaustive exploration of the topic.
It would be enough if even one of the identified threats is realized,
however, for user privacy to be seriously impacted. Also, it seems
probable that the risk will continue to grow with further
improvements of sensor technologies in terms of cost, size and
accuracy, further advances in machine learning methods, and
further proliferation of accelerometer-equipped mobile devices.
Given the widespread perception of accelerometers as non-
intrusive, we call for an urgent reconsideration of their privacy
implications, along with corresponding adjustments to technical
Figure 3: Overview of sensitive inferences that can be drawn from accelerometer data (according to the referenced studies).
84

and legal protection measures. In our opinion, the sensitivity of
sensor data should generally be assessed in consideration of all
inferences that could plausibly be drawn from it, and not based on
the sensor’s official purpose. Further research into the privacy-
intrusion potential of accelerometers and other seemingly benign
sensors is needed, taking into account state-of-the-art data mining
techniques. As it is extremely difficult, however, to meaningfully
determine the limits of continuously advancing inference
methods, most sensors in mobile devices should be regarded and
treated as highly sensitive by default.
4. CONCLUSION
Accelerometers are among the most widely used sensors in
mobile devices, where they have a large variety of possible
applications. They are commonly regarded as not privacy-
intrusive and therefore often less access-restricted than other
sensors, such as cameras and microphones. However, based on
existing literature, we found that accelerometer data can enable
serious privacy intrusions by allowing inferences about a device
holder’s location, identity, demographics, personality, health
status, emotions, activities and body features.
Any trait or behavior of a user that results in characteristic
movement patterns can potentially be detected through accelera-
tion signals. Accelerometers are cheap, low in power consumption
and often invisibly embedded into consumer devices. Thus, they
represent a perfect surveillance tool as long as their data streams
are not properly monitored and protected from potentially
untrusted parties such as device manufacturers, service providers
and app developers. In current mobile operating systems, third-
party apps can access accelerometer data without requiring any
permission or conscious participation from the user.
Although this paper conveys only a first impression of the privacy
violations that could be enabled through accelerometers, the
findings already are significant enough to express a warning to
consumers who could be affected, as well as a call for action to
the public and private actors who are entrusted with protecting
user privacy in mobile devices.
5. REFERENCES
[1] Wearable Devices That Have an Accelerometer: 2018.
https://vandrico.com/wearables/device-categories/
components/accelerometer. Accessed: 2018-06-06.
[2] Richardson, S. and Mackinnon, D. 2017. Left to their own
Devices? Privacy Implications of Wearable Technology in
Canadian Workplaces. Surveillance Studies Centre.
[3] Bai, X. et al. 2017. Sensor Guardian: prevent privacy
inference on Android sensors. EURASIP Journal on
Information Security. 2017, 1 (Dec. 2017).
DOI:https://doi.org/10.1186/s13635-017-0061-8.
[4] Hnat, T.W. et al. 2012. Doorjamb: unobtrusive room-level
tracking of people in homes using doorway sensors.
Proceedings of the 10th ACM Conference on Embedded
Network Sensor Systems (2012), 309–322.
[5] Klasnja, P. et al. 2009. Exploring privacy concerns about
personal sensing. International Conference on Pervasive
Computing (2009), 176–183.
[6] Bojinov, H. et al. 2014. Mobile device identification via
sensor fingerprinting. arXiv:1408.1416. (2014).
[7] Xu, Z. and Zhu, S. 2015. SemaDroid: A Privacy-Aware
Sensor Management Framework for Smartphones.
Proceedings of the 5th ACM Conference on Data and
Application Security and Privacy (2015), 61–72.
[8] Weiss, G.M. et al. 2016. Actitracker: A Smartphone-Based
Activity Recognition System for Improving Health and Well-
Being. IEEE International Conference on Data Science and
Advanced Analytics (DSAA) (Oct. 2016), 682–688.
[9] Matic, A. et al. 2013. Automatic Sensing of Speech Activity
and Correlation with Mood Changes. Pervasive and Mobile
Sensing and Computing for Healthcare. Springer Berlin
Heidelberg. 195–205.
[10] Crouter, S. et al. 2003. Validity of 10 Electronic Pedometers
for Measuring Steps, Distance, and Energy Cost. Med. Sci.
Sport Exer. 35, (2003), 1455–60.
[11] Migueles, J. et al. 2017. Accelerometer Data Collection and
Processing Criteria to Assess Physical Activity and Other
Outcomes: A Systematic Review and Practical
Considerations. Sports Medicine. 47, (2017).
[12] Yang, C.-C. and Hsu, Y.-L. 2010. A Review of
Accelerometry-Based Wearable Motion Detectors for
Physical Activity Monitoring. Sensors. 10, 8 (Aug. 2010),
7772–7788. DOI:https://doi.org/10.3390/s100807772.
[13] Chernbumroong, S. et al. 2011. Activity classification using
a single wrist-worn accelerometer. 2011 5th International
Conference on Software, Knowledge Information, Industrial
Management and Applications (SKIMA) Proceedings (Sep.
2011), 1–6.
[14] Gupta, P. and Dallas, T. 2014. Feature Selection and Activity
Recognition System Using a Single Triaxial Accelerometer.
IEEE Transactions on Biomedical Engineering. 61, 6 (Jun.
2014), 1780–1786.
DOI:https://doi.org/10.1109/TBME.2014.2307069.
[15] Khan, A.M. et al. 2010. A Triaxial Accelerometer-Based
Physical-Activity Recognition via Augmented-Signal
Features and a Hierarchical Recognizer. IEEE Transactions
on Information Technology in Biomedicine. 14, 5 (Sep.
2010), 1166–1172.
DOI:https://doi.org/10.1109/TITB.2010.2051955.
[16] Lee, J.-V. et al. 2013. Smart Elderly Home Monitoring
System with an Android Phone. International Journal of
Smart Home. 7, 3 (2013), 16.
[17] Mannini, A. et al. 2013. Activity recognition using a single
accelerometer placed at the wrist or ankle. Med. Sci. Sport
Exer. 45, 11 (Nov. 2013), 2193–2203.
DOI:https://doi.org/10.1249/MSS.0b013e31829736d6.
[18] Tapia, E.M. 2008. Using machine learning for real-time
activity recognition and estimation of energy expenditure.
Massachusetts Institute of Technology.
[19] Taraldsen, K. et al. 2012. Physical activity monitoring by use
of accelerometer-based body-worn sensors in older adults: A
systematic literature review of current knowledge and
applications. Maturitas. 71, 1 (Jan. 2012), 13–19.
DOI:https://doi.org/10.1016/j.maturitas.2011.11.003.
[20] Bhagat, Y.A. et al. 2014. Clinical validation of a wrist
actigraphy mobile health device for sleep efficiency analysis.
2014 IEEE Healthcare Innovation Conference (HIC) (Oct.
2014), 56–59.
[21] Pesonen, A.-K. and Kuula, L. 2018. The Validity of a New
Consumer-Targeted Wrist Device in Sleep Measurement: An
Overnight Comparison Against Polysomnography in
Children and Adolescents. J. Clin. Sleep Med. 14, 04 (Apr.
2018), 585–591. DOI:https://doi.org/10.5664/jcsm.7050.
[22] Liu, J. et al. 2009. uWave: Accelerometer-based
Personalized Gesture Recognition and Its Applications.
85

Pervasive Mob. Comput. 5, 6 (Dec. 2009), 657–675.
DOI:https://doi.org/10.1016/j.pmcj.2009.07.007.
[23] Thomaz, E. et al. 2015. A Practical Approach for
Recognizing Eating Moments with Wrist-Mounted Inertial
Sensing. UBICOMP. 2015, (Sep. 2015), 1029–1040.
DOI:https://doi.org/10.1145/2750858.2807545.
[24] Zhang, S. et al. 2009. Detection of Activities by Wireless
Sensors for Daily Life Surveillance: Eating and Drinking.
Sensors. 9, 3 (Mar. 2009), 1499–1517.
DOI:https://doi.org/10.3390/s90301499.
[25] Saleheen, N. et al. 2015. puffMarker: A Multi-Sensor
Approach for Pinpointing the Timing of First Lapse in
Smoking Cessation. UBICOMP. 2015, (Sep. 2015), 999–
1010.
[26] Tang, Q. 2014. Automated Detection of Puffing and
Smoking with Wrist Accelerometers. 8th International
Conference on Pervasive Computing Technologies for
Healthcare (2014).
[27] Killian, J. 2018. Smartphone-Based Intelligent System:
Training AI to Track Sobriety Using Smartphone Motion
Sensors. The Ohio State University.
[28] Gharani, P. et al. 2017. An Artificial Neural Network for
Gait Analysis to Estimate Blood Alcohol Content Level.
Computing Research Repository. (2017).
[29] Arnold, Z. et al. 2015. Smartphone Inference of Alcohol
Consumption Levels from Gait. 2015 International
Conference on Healthcare Informatics (Oct. 2015), 417–426.
[30] Williamson, J.R. et al. 2015. Estimating load carriage from a
body-worn accelerometer. BSN (Jun. 2015), 1–6.
[31] Singh, P. et al. 2013. Using Mobile Phone Sensors to Detect
Driving Behavior. Proceedings of the 3rd ACM Symposium
on Computing for Development (2013), 53:1–53:2.
[32] Vaiana, R. et al. 2014. Driving Behavior and Traffic Safety:
An Acceleration-Based Safety Evaluation Procedure for
Smartphones. Modern Applied Science. 8, (2014), 88–96.
[33] Dai, J. et al. 2010. Mobile phone based drunk driving
detection. 2010 4th International Conference on Pervasive
Computing Technologies for Healthcare (Mar. 2010), 1–8.
[34] Englebienne, G. and Hung, H. 2012. Mining for motivation:
using a single wearable accelerometer to detect people’s
interests. Proceedings of the 2nd ACM international
workshop on Interactive multimedia on mobile and portable
devices (2012), 23.
[35] Zhang, L. et al. 2015. AccelWord: Energy Efficient Hotword
Detection through Accelerometer. MOBISYS (2015), 301–
315.
[36] Dusan, S.V. et al. 2016. System and method of detecting a
user’s voice activity using an accelerometer. US9438985B2.
Sep. 6, 2016.
[37] Mohapatra, P. et al. 2017. Energy-efficient, accelerometer-
based hotword detection to launch a voice-control system.
US20170316779A1. Nov. 2, 2017.
[38] Han, J. et al. 2012. ACComplice: Location inference using
accelerometers on smartphones. Fourth International
Conference on Communication Systems and Networks (Jan.
2012), 1–9.
[39] Hua, J. et al. 2015. We Can Track You If You Take the
Metro: Tracking Metro Riders Using Accelerometers on
Smartphones. arXiv:1505.05958. (May 2015).
[40] Nickel, C. et al. 2012. Authentication of Smartphone Users
Based on the Way They Walk Using k-NN Algorithm.
Proceedings of the 2012 Eighth International Conference on
Intelligent Information Hiding and Multimedia Signal
Processing (2012), 16–20.
[41] Primo, A. et al. 2014. Context-Aware Active Authentication
Using Smartphone Accelerometer Measurements. IEEE
Conference on Computer Vision and Pattern Recognition
Workshops (Jun. 2014), 98–105.
[42] Srilekha, R. and Jayakumar, D. 2015. A Secure Screen Lock
System for Android Smart Phones Using Accelerometer
Sensor. International Journal For Science Technology And
Engineering. 1, 10 (May 2015), 96–100.
[43] Li, S. et al. 2016. Whose move is it anyway? Authenticating
smart wearable devices using unique head movement
patterns. IEEE International Conference on Pervasive
Computing and Communications (Mar. 2016), 1–9.
[44] Kwapisz, J.R. et al. 2010. Cell phone-based biometric
identification. IEEE International Conference on Biometrics:
Theory, Applications and Systems (Sep. 2010), 1–7.
[45] Freudiger, J. et al. 2011. Evaluating the privacy risk of
location-based services. International conference on
financial cryptography and data security (2011), 31–46.
[46] Golle, P. and Partridge, K. 2009. On the anonymity of
home/work location pairs. International Conference on
Pervasive Computing (2009), 390–397.
[47] Dey, S. et al. 2014. AccelPrint: Imperfections of
Accelerometers Make Smartphones Trackable. Network and
Distributed System Security Symposium (2014).
[48] Das, A. et al. 2016. Tracking Mobile Web Users Through
Motion Sensors: Attacks and Defenses. Network and
Distributed System Security Symposium (2016).
[49] Cai, L. and Chen, H. 2011. TouchLogger: Inferring
Keystrokes on Touch Screen from Smartphone Motion.
HotSec. 11, (2011), 9–9.
[50] Wang, H. et al. 2015. MoLe: Motion Leaks Through
Smartwatch Sensors. MOBICOM (2015), 155–166.
[51] Xu, Z. et al. 2012. Taplogger: Inferring user inputs on
smartphone touchscreens using on-board motion sensors.
Proceedings of the fifth ACM conference on Security and
Privacy in Wireless and Mobile Networks (2012), 113–124.
[52] Aviv, A.J. et al. 2012. Practicality of Accelerometer Side
Channels on Smartphones. Proceedings of the 28th Annual
Computer Security Applications Conference (2012), 41–50.
[53] Owusu, E. et al. 2012. ACCessory: Password Inference
Using Accelerometers on Smartphones. Proceedings of the
Twelfth Workshop on Mobile Computing Systems (2012),
9:1–9:6.
[54] Song, Y. et al. 2014. Two Novel Defenses against Motion-
Based Keystroke Inference Attacks. Computing Research
Repository. (2014).
[55] Beltramelli, T. and Risi, S. 2015. Deep-Spying: Spying using
Smartwatch and Deep Learning. arXiv:1512.05616. (Dec.
2015).
[56] Weiss, G.M. and Lockhart, J.W. 2011. Identifying user traits
by mining smart phone accelerometer data. Proceedings of
the Fifth International Workshop on Knowledge Discovery
from Sensor Data (2011).
86

[57] Yanai, H.-F. and Enjyoji, A. 2016. Estimating Carrier’s
Height by Accelerometer Signals of a Smartphone. HCI
International (Cham, 2016), 542–546.
[58] Ferrari, G.L. de M. et al. 2017. Accelerometer-determined
peak cadence and weight status in children from São Caetano
do Sul, Brazil. Ciência & Saúde Coletiva. 22, (Nov. 2017),
3689–3698. DOI:https://doi.org/10.1590/1413-
812320172211.21962015.
[59] Warburton, D.E.R. et al. 2006. Health benefits of physical
activity: the evidence. Can. Med. Assoc. J. 174, 6 (Mar.
2006), 801–809. DOI:https://doi.org/10.1503/cmaj.051351.
[60] Zeitzer, J.M. et al. 2018. Daily Patterns of Accelerometer
Activity Predict Changes in Sleep, Cognition, and Mortality
in Older Men. J. Gerontol. 73, 5 (Apr. 2018), 682–687.
DOI:https://doi.org/10.1093/gerona/glw250.
[61] Chan, C.B. et al. 2012. Cross-sectional Relationship of
Pedometer-Determined Ambulatory Activity to Indicators of
Health. Obesity Research. 11, 12 (2012), 1563–1570.
DOI:https://doi.org/10.1038/oby.2003.208.
[62] Alvarez, G.G. and Ayas, N.T. 2007. The Impact of Daily
Sleep Duration on Health: A Review of the Literature.
Progress in Cardiovascular Nursing. 19, 2 (2007), 56–59.
DOI:https://doi.org/10.1111/j.0889-7204.2004.02422.x.
[63] Hees, V.T. van et al. 2015. A Novel, Open Access Method to
Assess Sleep Duration Using a Wrist-Worn Accelerometer.
PLOS One. 10, 11 (Nov. 2015).
DOI:https://doi.org/10.1371/journal.pone.0142533.
[64] Borghese, M. et al. 2018. Estimating sleep efficiency in 10-
to- 13-year-olds using a waist-worn accelerometer. Sleep
Health. 4, 1 (Feb. 2018), 110–115.
DOI:https://doi.org/10.1016/j.sleh.2017.09.006.
[65] Lei, Z. et al. 2017. Supervised learning in voice type
discrimination using neck-skin vibration signals: Preliminary
results on single vowels. J. Acoust. Soc. Am. 141, 5 (May
2017), 3916–3916. DOI:https://doi.org/10.1121/1.4988844.
[66] Liu, C. et al. 2015. A physiological sound sensing system
using accelerometer based on flip-chip piezoelectric
technology and asymmetrically gapped cantilever. 2015
IEEE 65th Electronic Components and Technology
Conference (ECTC) (May 2015), 1874–1877.
[67] Smidt, G. et al. 1972. Accelerographic analysis of several
types of walking. Amer. J. Physical Med. 50, (1972), 285–
300.
[68] Menz, H.B. et al. 2003. Age-related differences in walking
stability. Age and Ageing. 32, 2 (Mar. 2003), 137–142.
DOI:https://doi.org/10.1093/ageing/32.2.137.
[69] Davarci, E. et al. 2017. Age group detection using
smartphone motion sensors. 2017 25th European Signal
Processing Conference (Aug. 2017), 2201–2205.
[70] Cho, S.H. et al. 2004. Gender differences in three
dimensional gait analysis data from 98 healthy Korean
adults. Clin. Biomech. 19, 2 (Feb. 2004), 145–152.
DOI:https://doi.org/10.1016/j.clinbiomech.2003.10.003.
[71] Jain, A. and Kanhangad, V. 2016. Investigating gender
recognition in smartphones using accelerometer and
gyroscope sensor readings. International Conference on
Computational Techniques in Information and
Communication Technologies (Mar. 2016), 597–602.
[72] Jago, R. et al. 2005. Adolescent patterns of physical activity
differences by gender, day, and time of day. Amer. J.
Physical Med. 28, 5 (Jun. 2005), 447–452.
DOI:https://doi.org/10.1016/j.amepre.2005.02.007.
[73] MERRIFIELD, H.H. 1971. Female Gait Patterns in Shoes
with Different Heel Heights. Ergonomics. 14, 3 (May 1971),
411–417. DOI:https://doi.org/10.1080/00140137108931260.
[74] Kanning, M. and Schlicht, W. 2010. Be Active and Become
Happy: An Ecological Momentary Assessment of Physical
Activity and Mood. Journal of Sport and Exercise
Psychology. 32, 2 (Apr. 2010), 253–261.
DOI:https://doi.org/10.1123/jsep.32.2.253.
[75] Gruenenfelder-Steiger, A.E. et al. 2017. Physical Activity
and Depressive Mood in the Daily Life of Older Adults. The
Journal of Gerontopsychology and Geriatric Psychiatry. 30,
3 (Jan. 2017), 119–129. DOI:https://doi.org/10.1024/1662-
9647/a000172.
[76] Zhang, Z. et al. 2016. Emotion recognition based on
customized smart bracelet with built-in accelerometer. PeerJ.
4, (Jul. 2016). DOI:https://doi.org/10.7717/peerj.2258.
[77] Garcia-Ceja, E. et al. 2016. Automatic Stress Detection in
Working Environments from Smartphones’ Accelerometer
Data: A First Step. IEEE Journal of Biomedical and Health
Informatics. 20, 4 (Jul. 2016), 1053–1060.
DOI:https://doi.org/10.1109/JBHI.2015.2446195.
[78] Olsen, A.F. and Torresen, J. 2016. Smartphone
accelerometer data used for detecting human emotions. 2016
3rd International Conference on Systems and Informatics
(ICSAI) (Nov. 2016), 410–415.
[79] Wilson, K. and Dishman, R. 2014. Personality and Physical
Activity: A Systematic Review and Meta-analysis. Med. Sci.
Sport Exer. (May 2014), 473.
[80] Artese, A. et al. 2017. Personality and Actigraphy-Measured
Physical Activity in Older Adults. Psychology and Aging. 32,
2 (Mar. 2017), 131–138.
DOI:https://doi.org/10.1037/pag0000158.
[81] Wilson, K.E. et al. 2015. Personality Correlates of Physical
Activity in College Women: Medicine & Science in Sports &
Exercise. 47, 8 (Aug. 2015), 1691–1697.
DOI:https://doi.org/10.1249/MSS.0000000000000570.
[82] Foerster, F. et al. 1999. Detection of posture and motion by
accelerometry: a validation study in ambulatory monitoring.
Computers in Human Behavior. 15, 5 (Sep. 1999), 571–583.
DOI:https://doi.org/10.1016/S0747-5632(99)00037-0.
[83] Khan, A.M. et al. 2010. Accelerometer’s position free human
activity recognition using a hierarchical recognition model.
The 12th IEEE International Conference on e-Health
Networking, Applications and Services (Jul. 2010), 296–301.
87