IEEE Consumer Electronics Magazine - September/October 2020 - 25

Table 1. Characteristics of IMU sensors.
Accelerometer

Gyroscope

Sampling rate

20 Hz

Dynamic range

Æ2 g

Æ125 dps

Æ1024 LSB/g
pﬃﬃﬃﬃﬃﬃ
150 mg= Hz

Æ262 LSB/dps
pﬃﬃﬃﬃﬃﬃ
0.014 dps= Hz

Sensitivity
Noise density

on-chip (BLE SoC) module. Each IPMS module is
developed using a small form-factor BLE SoC
board, to facilitate the communication between
the sensors and a mobile application. Regarding
the MEMS IMU, we utilized an off-the-shelf lowpower sensor able to meet the needs of lowpower operation required by the Bluetooth Low
Energy SoC. The specific characteristics of the
MEMS sensors (accelerometer and gyroscope)
are shown in Table 1.
Focusing on the IPMS mobile application, the
user initially participates in the placement calibration phase. During a session, the IPMS notifies the user using the vibrator and visual alerts
when the angle thresholds are violated indicating abnormal sitting posture.
Finally, after the session is completed, the
user can observe his posture behavior, while
these data are stored on a cloud-based back-end
for posture performance monitoring over time.9
The collected datasets can be further utilized
using machine learning techniques (deep learning, etc.) to feed health-oriented applications
and lead to disorder prognosis and early
detection.

EVALUATION OF IPMS
In this section, a detailed evaluation of the
proposed IPMS is presented through a custom
experimental procedure. Since the proposed
approach consists of individual modules for
sensing the spine misalignment, we tested and
evaluated the estimation of the two system
angles, u and f.
For the testing procedure an IPMS module
was attached on a PTU-D46 Pan/Tilt Unit.9 The
sensor's mounting resembles the placement on
the human back, while the servo motor moves to
access different predefined u and f angles,

September/October 2020

Figure 4. Root mean square error-angles under
rotation: (a) u, (b) f.

ranging from À10 to þ10 with 1 step. This setup
assumes that the user will use the IPMS devices in
an almost steady state-slouching or embracing
a good posture-for a period of time. Therefore,
we want to investigate the angle deviation in
terms of the root mean square error for a fixed
number of samples. During this time the esti^ yielded by the IPMS are
mated angles ð^u and fÞ
compared with the real ones ðu and fÞ accessed
by the servo motor with its predefined angle
resolution.
In order to evaluate the measurement accuracy for the u, f angles (see Figure 2) the servo
motor rotates the module by modifying one of
these angles at a time. This procedure is executed
in order to evaluate these predefined angles
while no rotation occurs at the other angle. For
each angle position, we collected N samples for
five consecutive minutes, and we repeated this
experiment for eight independent trials. For each
trial, we computed the root mean square error
between the actual
and the estimated angle (^u or
qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
P
2
^ i.e., RMSE ¼ 1 N ð^
f),
n¼1 an À aÞ .
N
Figure 4(a) shows the average root mean
square error along with the standard deviation for the u angle among all trials, while
Figure 4(b) depicts the error for the f angle in
a similar manner. To highlight the estimated
angle error distribution, we present the histograms of u, f angles for 0 and 10 , respectively. As seen in Figure 5, the angles' error
distributions exhibit a Gaussian distribution
shape.

IEEE Consumer Electronics Magazine - September/October 2020

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - September/October 2020