Background - Smartphone-Based Human Activity Recognition - page 12

the surroundings, the users that are part of it, and their interaction. There is a wide
range of ambient sensors such as microphones, video cameras, presence sensors,
thermometers and depth sensors.
Many of these sensors have already been employed for HAR. For example, video
cameras are employed in Poppe ( 2007 ) for markerless vision-based human motion
analysis. Also in Bian et al. ( 2005 ), microphones are used for sound source local-
ization in a home environment for communication activity inference. Even everyday
use devices can also indicate the occurrence of certain activities such as the identifi-
cation of switched on/off household appliances to infer home activities (Ogawa et al.
2002 ). Moreover, in Takaˇcetal.( 2013 ), a Microsoft Kinect's depth sensor is used
as an ambient sensor for position and orientation tracking for an indoor monitoring
system for Parkinson's disease (PD) patients. Although these examples can provide
accurate context information about the agent's motion and localization, they require
a static infrastructure which limits its range of operation to a constrained space.
Additionally, video-based systems can be very effective for HAR but they are
somewhat disadvantageous due to their demanding computations (e.g. for achiev-
ing real-time operation) and privacy concerns, for instance, when they are used in
home environments as people are generally uncomfortable about being continuously
monitored.
2.3.2 Wearable Sensors
Wearable sensors are used to gather signals directly from users. They are commonly
attached to different body parts such as the waist, wrist, chest, legs and head (Ravi
et al. 2005 ) but also fit to clothes and embedded in other accessories of regular use such
as watches, glasses or mobile phones (Brezmes et al. 2009 ). They contain a battery
unit that provides the energy supply for continuous operation and, some of them,
also have a wireless unit for sensor data transmission when it is externally needed or
to interface them with other body-worn devices. Physiological and motion signals
obtained from wearable sensors are highly informative for HAR. Skin temperature,
heart rate, heat flux, conductivity, Global Positioning System (GPS) location and
body motion are some examples of variables that can be measured with current
wearable sensor technologies (Yang and Yacoub 2006 ). These can be convenient,
for example, in healthcare applications where it is possible to exploit them for the
continuous monitoring of patients and the detection of an emergent health condition.
Contrarily to ambient sensors, wearable sensors have advantages regarding pri-
vacy and area of operation. In the first case, users are less reluctant to use them in
every location if there is no capture of images or video. Secondly, considering that
these sensors are always carried by the user, they are ubiquitous and their location
coverage is virtually unconstrained. They have also the advantage of being highly
portable and do not require fixed equipment. On the other hand, wearable sensors