IEEE Systems, Man and Cybernetics Magazine - October 2020 - 11

Posture and Activity Recognition
Activity Recognition Based on Adaptive
Sliding Windows
Among the elderly population and people who suffer
from motor disabilities, there is often prolonged use of
wheelchairs, with health risks including chronic back
pain and pressure ulcers. The smart cushion was used
to detect the following basic seated postures: proper sitting (PS), LL, LR, LF, and lean backward (LB). We implemented the machine-learning method based on DT in
the Arduino platform.
We can distinguish different activity categories using
combinations of the postures, examples of which are provided in Table 3 and include the following:
◆ Stable posture activities are postures that are kept for
a meaningful amount of time. A PS activity is exemplified in Figure 3(a).
◆ Transitional posture activities consist of a sequence of
different individual sitting postures, such as the pressure relief activity useful in wheelchair users to prevent pressure ulcers. A relevant example of left
pressure relief activity is depicted in Figure 3(b).
◆ Noncumulative posture activities refer to sporadic sitting actions (see Figure 4) and include trunk exercises,
which could strengthen core stability and promote the
prevention of several medical complications such as
chronic strokes [48].
In previous research [49], we proposed an adaptive sliding window (ADW)-based classification algorithm focused
on the activities listed in Table 3. We collected data from
several subjects using the smart cushion deployed on a
wheelchair in static and dynamic conditions.
We implemented the algorithm in MATLAB. DT and
kNN were chosen to test our algorithm. In both static and
dynamic conditions, our algorithm obtained a higher
F-score (a combined metric of precision and recall). Table 4
lists the results of static condition between fixed-size windows and ADWs.

Activity Level Assessment for Sedentary
Lifestyle Users
It is well known that a sedentary lifestyle is strongly correlated with several health issues, therefore, medical
guidelines agree on the importance of constant motivation to maintain an active lifestyle. Activity level assessments can provide statistical information in a given
period and recommend that the users increase physical
exercise. In a previous study [44], we proposed an activity
level assessment system specifically conceived for wheelchair users. The architecture of the system, displayed in
Figure 5, consists of two layers:
◆ data preprocessing: time series of sampled data are
collected and instance vectors are generated from the
pressure sensors and IMU.
◆ activity assessment index calculation: significant features are extracted and machine-learning methods are
used to recognize the activity levels. Three activity levels are identified and categorized according to their
intensity, as summarized in Table 5.

0°
PS

30 -~60°
TTL

0°
PS

30 -~60°
TTR

0°
PS

Figure 4. Noncumulative-posture activity examples

(trunk-rotation exercise).

Table 3. The activity categories and their
examples.
Activity Category

Activity Example

Stable posture activity

PSA, LSA, RSA

Transitional posture activity

LPR and RPR

Noncumulative posture activity

TTL and TTR

PSA: proper sitting activity; LSA: left sitting activity; RSA: right sitting activity; LPR:
left pressure relief; RPR: right pressure relief; TTL: trunk turn left; TTR: trunk turn right.

(a)
PS

LL
(b)

Figure 3. Examples of cumulative posture activity.

(They were composed by a series of postures listed
in the boxes.) (a) Posture-stable activity and
(b) posture-transitional activity.

Table 4. The results of F-score using the
two time window selection methods in
a static environment.
PSA

LSA

RSA

LPR

RPR

TTL

TTR

DT and FW

0.93

0.81

0.68

0.72

kNN and FW

0.93

0.91

0.7

0.84

0.67

0.76

DT and AW

0.94

0.95

0.94

0.92

0.86

0.87

0.85

kNN and AW

0.95

0.93

0.94

0.89

0.87

0.82

0.84

FW: fixed-size windows.

O c tob e r 2020

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

IEEE Systems, Man and Cybernetics Magazine - October 2020

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