IEEE Systems, Man and Cybernetics Magazine - October 2020 - 7

Activity Recognition
In the context of body area networks (BANs) [1], many
wearable devices have emerged in recent years [2], [3]
that are changing people's lifestyles in various ways,
such as health care, fitness, sports [4], and entertainment. Along with multisensor data fusion techniques [5],
many kinds of useful information have been extracted
and valuable decisions can be made to better guide daily
life activities.
Initially, activity recognition was addressed with
vision-based approaches that introduced relevant privacy concerns; additionally, video-based approaches are
significantly more power consuming and require complex processing algorithms. Therefore, research interest gradually shifted toward wearable motion sensors,
thanks to their noninvasiveness, limited cost, and
practical implementation. The most popular wearable
sensor for activity recognition is the inertial measurement unit (IMU) [6], especially for its accelerometer.

Different features were extracted, and diversified
machine-learning methods were used to better recognize physical activities and static postures. For dynamic activities such as walking, running, jumping, or
falling, accelerometer sensors offer advantages. However, for sitting postures [7], sleeping activity [8], or some
gait posture recognition [9], pressure sensor-based
devices are actually more suitable than IMUs, as they
are easy to implement and are often deployed on
objects the user has not worn, whose iconic example is
the smart cushion.
This article provides a comprehensive overview of
activity recognition and human-machine interaction
applications based on the pressure-sensing smart cushion
and particularly reviews our related previous research
studies and results. Different from the Intelligent Wheelchair, our article is the first to focus on a smart cushionbased chair/wheelchair that was used to recognize
sitting-related activities.
Literature Review of Sitting Posture
Recognition Using Smart Cushions
An important part of the global population (e.g., desk
workers, truck and bus drivers, and wheelchair-bound
users) spends most of their daily time in a seated condition, resulting in a sedentary lifestyle. Smart devices
used in BANs, such as the smart cushion, appear to
better understand individuals' sitting-related activities
while in standard chairs or wheelchairs [10], [11].
Smart cushions are realized using two alternative
technologica l approaches: dense pressure sensor
arrays or sparse pressure sensor arrays. In the following section, a state-of-the-art review for each approach
is reported.

©SHUTTERSTOCK.COM/SOLAR22

Smart Cushions With Dense Pressure
Sensor Arrays
Many studies have adopted e-textile solutions that can
contain more than 1,000 sensor units. Xu et al. [12] proposed a textile-based sensing system: the pressure sensor d at a were repre s ent ed by bi n a r y pre s su re
distribution data and then converted to gray-scale images for posture recognition. Also, a sensor matrix system
composed of more than 2,000 pressure sensors and
commercialized by Tekscan [13] was adopted by several
studies on sitting posture monitoring. Tan et al. [14]
used the Tekscan to analyze sitting pressure distribution in the form of gray-scale image maps. Mota and
Picard [15] also employed the Tekscan, proposing a
three-layer feedforward neural network to recognize
nine different sitting postures. Meyer et al. [16] developed a capacitive textile pressure sensor array cushion
and proposed the naïve Bayes classification method to
recognize 16 sitting postures. Liu et al. [17] developed a
smart cushion with 32 × 32 pressure sensors; a convolutional neural network (CNN)-based algorithm was used
O c tob e r 2020

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

http://www.SHUTTERSTOCK.COM/SOLAR22

IEEE Systems, Man and Cybernetics Magazine - October 2020

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - October 2020