Signal Processing - March 2016 - 84

Richness

Visible Video
Thermal Video
Smartphone*
Audio
Hand-Worn**
Wattmeter
Pressure
Magnetic Switch
Vital-Sign Monitoring
Vibration
Ultrasonic
Photoelectric

Wearable
Nonwearable

Infrared Motion

Perceived Privacy
*Gyroscope, Accelerometer, Microphone
**Accelerometer, Heart Rate, Step Count, Magnetic Field

figure 2. The richness of the sensors versus a user's perceived privacy.

Various types of hand-worn sensors are also considered in
many activity recognition scenarios. These types of sensors
include multifunctional wristwatches, magnetic sensors and
other types of bracelets. Accelerometers are often integrated
into wristwatches providing hand and arm gesture recognition capabilities [56]. In [57] an wrist-worn activity detector
was used to perform sleep/awake activity classification. In [58]
inertial sensors, accelerometers, and tilt switches were combined in a wrist-worn sensing unit to model users' rhythms
and as a result recognize daily activities. Hand-worn magnetic
sensors are able to distinguish between magnetic fields emitted by different electrical devices and recognize the activity
of a user [59]. Emergency buttons are often integrated into a
wristwatch and used to request help in case of an emergency
situation [31]. Also, step counters are often integrated into
wristwatches [57].
The modern smartphone offers a wealth of sensors, and
can further be used as a communication platform [60]. Usually, accelerometers, gyroscopes, a global positioning system, a magnetometer and a microphone are incorporated
into a modern smartphone device, providing all necessary
information for movement, action and activity recognition
including fall detection [61].
RFID tags are often used to detect the interaction of a person with an object and infer an ADL. In [62], RFID tags were
deployed on various kitchen utensils such as bowls, cutlery,
dishes, and jars to detect food preparation, eating, and drinking
as well as on various cupboards, the TV, and furniture. Similar
setups are considered in [63]-[65]. Matic et al. [66] focus specifically on monitoring dressing activity and detecting dressing failures. Often, RFID sensors are used in combination with
other sensors such as accelerometers [66].
84

In addition to these sensors, there is also a large variety
of sensors that monitor vital signs such as blood glucose,
humidity and temperature, blood pressure, heart rat, pulse
oximetry, CO2 gas, electrocardiography, electroencephalography, electromyography, and electrooculography. These
allow for the monitoring of a large set of human vital statistics and support activity recognition and ADL classification
tasks [67], [68].
The ability of all of these sensors to provide rich information about people's lives and biometrics can raise severe
privacy concerns. Figure 2 illustrates the richness of the
sensors versus the perceived privacy of a person using these
sensors. It is clear from the figure that sensors that provide
rich information about a person are usually not perceived as
privacy-preserving. For example, a video camera that allows
for the recognition of almost any human activity in its field
of view cannot be used in most rooms due to heavy privacy
violations. In contrast, magnetic switches can be placed in
every room without severely violating privacy, but they do
not provide exhaustive information on every human activity.
This can be partly improved by installing multiple instances
of low-informative sensors (e.g., magnetic switches) so that
richer insight into the user's activities can be achieved up
to the level of the most informative sensors. However, this
comes at the cost of increased installation efforts and complicates the deployment process.

State of the art in ADL experimental setups
A primary goal of AAL is to assess the self-maintenance of
elderly people still living at home. Therefore, many studies in
ambient intelligence focus on automatically recognizing
human activities that correspond to ADLs, such as bathing,

IEEE SIgnal ProcESSIng MagazInE

March 2016

Table of Contents for the Digital Edition of Signal Processing - March 2016