Hindawi

Journal of Sensors
Volume 2017, Article ID 6747921, 14 pages
https://doi.org/10.1155/2017/6747921

Research Article
Estimation of Individual Muscular Forces of the Lower Limb
during Walking Using a Wearable Sensor System

Suin Kim, Kyongkwan Ro, and Joonbum Bae

Department of Mechanical Engineering, UNIST, Ulsan, Republic of Korea

Correspondence should be addressed to Joonbum Bae; [email protected]

Received 28 October 2016; Revised 24 February 2017; Accepted 26 March 2017; Published 22 May 2017

Academic Editor: Jaime Lloret

Copyright © 2017 Suin Kim et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Although various kinds of methodologies have been suggested to estimate individual muscular forces, many of them require a costly
measurement system accompanied by complex preprocessing and postprocessing procedures. In this research, a simple wearable
sensor system was developed, combined with the inverse dynamics-based static optimization method. The suggested method can
be set up easily and can immediately convert motion information into muscular forces. The proposed sensor system consisted of
the four inertial measurement units (IMUs) and manually developed ground reaction force sensor to measure the joint angles and
ground reaction forces, respectively. To verify performance, the measured data was compared with that of the camera-based motion
capture system and a force plate. Based on the motion data, muscular efforts were estimated in the nine muscle groups in the lower
extremity using the inverse dynamics-based static optimization. The estimated muscular forces were qualitatively analyzed in the
perspective of gait functions and compared with the electromyography signal.

1. Introduction differences among individuals, and surrounding conditions.

Additionally, only superficial muscles can be measured and it
Many studies have been conducted to estimate muscular is difficult to block interference from neighboring muscles.
forces in living things, especially humans. To understand Moreover, the true relationship between EMG signals and
the mechanism of the human body performing a motor muscular forces is not yet fully understood. Other researchers
task quantitatively, it is essential to estimate muscular forces, have attempted to measure muscle hardness using ultrasound
which has many potential applications: (1) fundamental waves and muscle volume using strain sensors, but these have
and quantitative analysis of human motion [1], (2) design similar drawbacks to EMG [10, 11].
of systemic training and analysis of injury mechanisms in Recently, computational simulation programs, such as
the field of sports science [2], (3) diagnosis of abnormal OpenSim [12], were developed to analyze motor tasks. They
muscular functions and evaluation of the therapeutic effects provide a model-based estimation method, which enables
of rehabilitation [3], and (4) design and development of to build a musculoskeletal model and analyze movements
prosthetic and orthotic [4] and human-robot interaction and muscle functions conveniently. The fundamental algo-
systems such as an exoskeleton [5–7]. rithm within the simulation is “computed muscle control”
However, although various methodologies have been pro- (CMC), which aims to find the optimal muscle activation
posed, estimation of muscular forces remains as a challenging according to given kinematic data of the body, such as
task. Among them, the most simple is to connect a force joint angles, velocities, and acceleration [13]. The algorithm
transducer directly to a muscle [8]. However, it cannot be demands relatively high computational costs due to the
applied to the human because it requires surgery to the intrinsic calculation processes including both optimization
body. Surface electromyography (sEMG) is the most widely and forward dynamics.
accepted noninvasive method to measure the activation level The inverse dynamics-based static optimization is one
of a muscle [9]. However, it requires complex and frequent of model-based estimation methods, which requires little
calibration processes due to the high noise/signal ratio, computation cost, because it does not include a forward
2 Journal of Sensors

dynamic process. In addition, unlike the other model-based

estimation methods, it is possible to be implemented in real-
time, since the inverse dynamics-based static optimization
does not demand the full set of motion data in advance. The
inverse dynamics-based static optimization has been applied
to various motor tasks: walking [14–21], spinal compression
[22–27], finger [28], wrist [29], elbow [30, 31], arm, and IMU
shoulder [32–34] movements.
More specifically, the first step in the inverse dynamics-
based static optimization is to capture the motions and
external forces applied to the body. In previous research, the
body movements were usually measured by a camera-based
motion capture system, and external forces were measured by
a force plate. Because the measurement systems have inherent
limitations in terms of mobility, this is only available within
GRF sensor
a laboratory environment and natural movement may be
restricted by use of a fixed force plate. In addition, it is not
easy to equip the experimental instruments because of spatial
and cost limitations. Figure 1: Configuration of the wearable sensor system.
To overcome the limitations, the camera-based motion
capture system has been replaced with the inertial mea-
surement units (IMUs). Recently, full body motion capture
motion capture system with a force plate and the estimated
systems using the IMUs have been developed [35–39], and
muscular efforts were confirmed with the EMG signals.
there are several commercialized products by Xsens [40],
This paper is organized as follows: detailed procedures for
Perception Neuron [41], Synertial [42], Meta Motion [43],
the estimation of individual muscular forces are introduced
Technaid [44], and so on. Using the sensor system, it has
in the next section, according to the sequence of inverse
been tried to estimate biomechanical factors such as joint
dynamics-based static optimization. In detail, the inverse
moments of the lower limb [45–47]. However, the previous
dynamic equations of the lower extremity are derived; the
researches have not focused on combining the mobile motion
required body segment parameters are specified to solve the
capture system and a shoe-type ground reaction force sensor
problems. The static optimization problem is defined based
to investigate individual muscular efforts.
on the musculoskeletal model with selected muscle groups. In
Therefore, in this study, the muscular forces are inves-
the next section, the configuration and functions of the wear-
tigated by integrating a simple wearable sensor system and
able sensor system are described, followed by experimental
the inverse dynamics-based static optimization. The wearable
verification of its performance. In the Results and Discussion,
sensor consists of four IMUs and an insole-type ground
the experimental results are presented, including estimated
reaction force (GRF) sensor. Analogous to the proposed
joint moments and individual muscular forces of the lower
sensor system, the wireless M3D gait analysis system was
limb during normal gait.
developed by Tec Gihan and Human Dynamics Analy-
sis (HDA) was also developed by Insenco [46]. The two
systems are composed of a number of IMUs and force 2. Materials and Methods
transducers attached below the shoes. Even though a rigid
force transducer usually allows great accuracy of force 2.1. The Wearable Sensor System. The wearable sensor system
measurement and multiaxial application, the user’s natural consists of a wearable structure, four IMUs, and a manually
motion can be disturbed due to rigidity and weight of the developed insole-type GRF sensor (Figure 1). The objective
sensor. The foot plantar pressure sensor should be thin of the sensor system was to measure the rotational angles
and flexible and light-weight, less than 300 g, for natural of the hip, knee, and ankle joint and two-dimensional pelvic
gait of the user [48, 49]. The proposed GRF sensor was acceleration and GRF, not restraining the motion of the user.
made of silicone tubes covered with fabrics to be light and For that purpose, the IMU was better than a conventional
soft. rotary sensor, such as an encoder, which was required to
Based on the motion information collected by the be aligned with the articular axis. Furthermore, the GRF
wearable sensor system, the inverse dynamics-based static sensor was fabricated with silicone tubes attached on the
optimization method was applied to estimate the individual insole, which made it lighter and softer than typical force
muscular forces of the lower limb during walking. In our pre- transducers.
vious work, a prototype of the sensor system was developed
and preliminary experimental results were acquired [50]. 2.1.1. Inertial Measurement Unit (IMU). To measure joint
However, it was difficult to validate the estimated muscular angles, the four commercially available 9 DOFs IMUs were
efforts due to uncertainties of the measurement system itself. adopted, which provide Euler angles [51]. Since the Euler
Therefore, in this study, joint angles and GRF measured by angles were given with respect to the absolute coordinate
the proposed system are compared with that of vision-based system (i.e., the earth), the reference frame was defined to
Journal of Sensors 3

z
tr
R
x
x
Rt y
x
t e z
z z
x t t
s e
z
Rs x
z
x s s

(a) (b)

Projection onto
the z-x plane
(sagittal plane)

z z
x t tr x
휃k =
휃1 −
휃2
tr
s

휃1

휃2
z
x

z
휃k

(c) (d)

Figure 2: Procedures for converting the Euler angles from the IMUs to the knee joint angle: (a) extracting the rotation matrices from each
IMU, (b) calculating the directional vector of each segment, (c) transforming the reference coordinate from the earth to the trunk, and (d)
projecting the vectors into the sagittal plane and calculating the joint angle.

measure joint angles by attaching an IMU on the trunk. where 𝑐 and 𝑠 are cosine and sine and 𝛼, 𝛽, and 𝛾 are roll, pitch,
Therefore, the postures of each segment were calculated and yaw angles, respectively. Then, the longitudinal direction
as a unit vector with respect to the reference coordinate. of the segment was expressed with respect to the earth by
For example, the following provides the detailed steps for multiplying (2) to (1) (Figure 2(b)):
calculating knee joint angle (Figure 2). The roll axes of the
𝑐𝛾𝑐𝛽
two IMUs attached on the thigh and shank were aligned with
[𝑠𝛾𝑐𝛽]
the longitudinal axes of each segment (Figure 2(a)), setting V𝑠𝑒 =[ ], (3)
the approaching vectors along the longitudinal axis of each
[ −𝑠𝛽 ]
IMU:
𝑠 where the superscript 𝑒 and the subscript 𝑠 refer to the earth
V𝑠𝑠 = [(1, 0, 0)𝑇 ] , (1) and the segment.
where the superscript indicates the reference coordinate Next, the approaching vectors of the segments can be
where the vector is expressed and the subscript indicates expressed with respect to the reference coordinate of the IMU
the approaching vector under consideration out of the thigh, on the trunk by multiplying the inverse of (2) on the trunk
shank, and foot. Consequently, a rotational matrix of the IMU (Figure 2(c)):
was given as follows: V𝑠tr
𝑐𝛾𝑐𝛽 𝑐𝛾𝑠𝛽𝑠𝛼 − 𝑠𝛾𝑐𝛼 𝑐𝛾𝑠𝛽𝑐𝛼 + 𝑠𝛾𝑠𝛼 (𝑐𝛽1 𝑐𝛽2 𝑐 (𝛾1 − 𝛾2 ) + 𝑠𝛽1 𝑠𝛽2 )
[𝑠𝛾𝑐𝛽 𝑠𝛾𝑠𝛽𝑠𝛼 + 𝑐𝛾𝑐𝛼 𝑠𝛾𝑠𝛽𝑐𝛼 − 𝑐𝛾𝑠𝛼] [ ] (4)
𝑅=[ ], (2) =[
[ (−𝑐𝛽 𝑠𝛽 𝑠𝛼
1 2 1 + 𝑐𝛽2 (𝑐 (𝛾1 − 𝛾2 ) 𝑠𝛽1 𝑠𝛼1 − 𝑐𝛼1 𝑠 (𝛾1 − 𝛾2 )))]
],
[ −𝑠𝛽 𝑐𝛽𝑠𝛼 𝑐𝛽𝑐𝛼 ] [ (−𝑐𝛽1 𝑐𝛼1 𝑠𝛽2 + 𝑐𝛽2 (𝑐𝛾1 𝑐 (𝛾1 − 𝛾2 ) 𝑠𝛽1 + 𝑠𝛼1 𝑠 (𝛾1 − 𝛾2 ))) ]
4 Journal of Sensors

where the subscript “1” refers to the angles of the IMU on the
target segment and “2” represents the reference frame on the
trunk.
Finally, the direction vectors of the thigh and shank were
projected onto the sagittal plane, so that angle of the segments
with respect to the body reference frame was calculated using (b)
simple trigonometrical functions (Figure 2(d)):

𝑎
𝜃𝑖 = − sgn (𝑎) arccos ( ) (𝑖 = 1, 2, 3) , (5)
√𝑎2 + 𝑐2

where 𝜃 is the angle of the segment with respect to the

coordinate of the IMU on the trunk and 𝑎 and 𝑐 are the
first and third components of (4), respectively. The knee joint
angle 𝜃𝑘 was calculated as 𝜃1 − 𝜃2 .
An important issue was to align the IMU carefully along
the longitudinal direction of the segment; 𝑥𝑧 plane of each
IMU (in Figure 2) was set to be parallel to the sagittal plane
of the corresponding segment. To align each IMU, the subject (a) (c)
was asked to stand still, and the IMUs were attached on the
each segment maintaining the condition that the direction Figure 3: Manually developed GRF sensor with four air bladders:
(a) entire view, (b) pressure sensor module, and (c) air bladders
of gravity measured by the IMU was on 𝑥𝑧 plane. If the
attached to the insole.
condition was satisfied, the 𝑥𝑧 plane and the sagittal plane
could be considered parallel. The IMUs were attached to the
wearable structure which was made of straps and rubber
bands. The wearable structure hold the IMUs tightly in place which was associated with the amount of air inside and the
to ensure that alignment of IMUs was unchanged during geometry of the bladder. The coefficient of determination, 𝑅2 ,
experiments. The initial angles were set to zero before the start was also calculated, which indicated how the acquired data fit
of an experiment, by asking the subject to stand and make the a linear model. On the assumption that the slope remained
leg be straightened. All the same procedures were applied to unchanged, it was necessary to initialize the signal to zero for
calculate the joint angles at the hip and ankle. accurate measurements.

2.1.2. Ground Reaction Force (GRF) Sensor. The proposed 2.2. Inverse Dynamics-Based Static Optimization. Individual
GRF sensor was a modified version of Smart Shoes whose muscular forces of the lower limb were estimated by applying
performance was validated in the previous studies [52, 53]. inverse dynamics-based static optimization through a series
Figure 3 describes its configuration and the weight of the of procedures (Table 1). The lower extremity was modeled as a
entire insole sensor was 125 g. The GRF sensor only measured two-dimensional musculoskeletal system in the sagittal plane
force along the vertical axis from the ground by measuring that includes three segments (thigh, shank, and foot), three
pressure changes within the air bladders. The tangential GRF joints (hip, knee, and ankle), and nine muscle groups. The
was assumed to be negligible, because the magnitude of the model includes information on body segment parameters
vertical force is about 20 times more than both the lateral to solve the inverse dynamics problem and moment arms
shear force and progressional shear force during walking and the physiological cross-sectional area of each muscle
[54, 55]. From the force distribution of the four air bladders, for defining the static optimization problem. Based on the
the center of pressure (COP) of the GRF was also calculated musculoskeletal model, a set of motion data were measured,
with respect to the position of the ankle joint. the motion data were substituted into the inverse dynamic
It was necessary to determine the relationship between equations to estimate joint moments, and the individual mus-
the pressure changes within the air bladder and external cular forces were estimated by solving the static optimization
force applied to it. For the calibration process, while the air based on the joint moments.
bladder was pressurized, the corresponding external force
was measured using a force transducer while simultaneously 2.2.1. Inverse Dynamics. The lower extremity was described
measuring the air pressure within the air bladder. Before the as a mechanical system in dynamic equilibrium under the
calibration process, to remove hysteresis effect caused by the actions of gravity, inertia, ground reactions, and muscular
viscoelastic properties of the silicone tubes, dynamic com- forces, which generate appropriate moments at the joints.
pensator was applied to compensate dynamic effect of the air Figure 5 shows a free body diagram of the lower extremity in
bladders. The performance of the dynamic compensator was the sagittal plane, considering only flexion/extension motion
verified in the previous research [56, 57]. After the dynamic of the three joints, because most of the movements and
compensator was applied, the calibration procedures were forces in the lower limb during walking occur in the plane
conducted. The collected results were fitted with a line, as [58]. The two-dimensional model involves five degrees of
shown in Figure 4, to make a linear model, the slope of freedom, including three joint angles and the planar motion
Journal of Sensors 5

Bladder 1 Bladder 2
150 200
1 Slope = 1.8559 Slope = 1.1593
R-square = 0.9949 R-square = 0.9990
2 150
100
3

Force (N)

Force (N)
100

50
50
4
0 0
0 20 40 0 50 100 150
Pressure (mbar) Pressure (mbar)

Bladder 3 Bladder 4
100 150
Slope = 1.9692 Slope = 2.9041
R-square = 0.9870 R-square = 0.9206
100

Force (N)
Force (N)

50

50

0 0
0 10 20 30 40 0 10 20 30
Pressure (mbar) Pressure (mbar)

Figure 4: Results of the calibration process with the resulting slope and 𝑅2 value.

of the pelvis. Using Lagrange mechanics, the inverse dynamic + (𝑙𝑡2 (𝑚𝑓 + 𝑚𝑠 ) + 𝑑𝑡2 𝑚𝑡 ) 𝜃1̈ − 𝑑𝑠 𝑙𝑡 𝑚𝑠 𝜃2̈ cos (𝜃1 + 𝜃2 )
equations were derived with respect to the absolute angle of
the three segments: + 𝑑𝑓 𝑙𝑡 𝑚𝑓 𝜃3̈ cos (𝜃1 − 𝜃3 ) + (𝑙𝑡 (𝑚𝑓 + 𝑚𝑠 ) + 𝑑𝑡 𝑚𝑡 )

𝑀𝐴 = −𝐹GRF 𝑟 sin 𝜃3 + 𝐽𝑓 𝜃3̈ + 𝑔 sin 𝜃3 ⋅ 𝑥𝑝̈ cos 𝜃1 + (𝑙𝑡 (𝑚𝑓 + 𝑚𝑠 ) + 𝑑𝑡 𝑚𝑡 ) 𝑦𝑝̈ sin 𝜃1 ,
(6)
+ 𝑑𝑓 𝑚𝑓 (−𝑙𝑡 𝜃12̇ sin (𝜃1 − 𝜃3 ) + 𝑙𝑠 𝜃22̇ sin (𝜃2 + 𝜃3 )
where the capital subscripts indicate the relevant joint, hip,
+ 𝑙𝑡 𝜃12̈ cos (𝜃1 − 𝜃3 ) + 𝑙𝑠 𝜃2̈ cos (𝜃2 + 𝜃3 ) + 𝑑𝑓 𝜃3̈ knee, and ankle and the small letters represent the segment:
pelvis, thigh, shank, and foot. 𝑀 is the joint moment
generated by the muscles, 𝑑 is the center of mass positions
+ 𝑥𝑝̈ cos 𝜃3 + 𝑦𝑝̈ sin 𝜃3 ) ,
from the proximal joints, 𝜃1 , 𝜃2 , and 𝜃3 are the absolute angles
of the thigh, shank, and foot, 𝑥 and 𝑦 are the center of mass
𝑀𝐾 = 𝑀𝐴 + 𝑙𝑠 𝐹GRF sin 𝜃2 − 𝐽𝑠 𝜃2̈ − 𝑔 (𝑙𝑠 𝑚𝑓 + 𝑑𝑠 𝑚𝑠 ) position of each segment, 𝐹GRF is the ground reaction, and 𝑟
is the application point of the ground reaction from the ankle.
⋅ sin 𝜃2 − 𝑙𝑡 (𝑙𝑠 𝑚𝑓 + 𝑑𝑠 𝑚𝑠 ) 𝜃12̇ sin (𝜃1 + 𝜃2 ) Except for the measured motion data, there exist some
required parameters to solve the dynamic equations such
− 𝑑𝑓 𝑙𝑠 𝑚𝑓 𝜃32̇ sin (𝜃2 + 𝜃3 ) + (𝑙𝑠 𝑚𝑓 + 𝑑𝑠 𝑚𝑠 ) 𝑙𝑡 𝜃1̈ as segment length, mass, center of mass position, and mass
moment of inertia. However, they are not directly measur-
⋅ cos (𝜃1 + 𝜃2 ) − (𝑙𝑠2 𝑚𝑓 + 𝑑𝑠2 𝑚𝑠 ) 𝜃2̈ + 𝑑𝑓 𝑙𝑠 𝑚𝑓 𝜃3̈ able, except for the segment length; thus the other parameters
should be estimated using the formulas found in cadaveric
⋅ cos (𝜃2 + 𝜃3 ) + (𝑙𝑠 𝑚𝑓 + 𝑑𝑠 𝑚𝑠 ) 𝑥𝑝̈ cos 𝜃2 − (𝑙𝑠 𝑚𝑓 experiments [59]:
𝑚 = 𝑃 ⋅ (body mass) ,
+ 𝑑𝑠 𝑚𝑠 ) 𝑦𝑝̈ sin 𝜃2 ,
𝑐 = 𝑅 ⋅ (segment length) , (7)
𝑀𝐻 = 𝑀𝐾 − 𝑙𝑡 𝐹GRF sin 𝜃2 + 𝐽𝑡 𝜃1̈ + 𝑔 (𝑙𝑡 (𝑚𝑓 + 𝑚𝑠 ) 2
𝐼 = 𝑚 ⋅ 𝐾 ⋅ (segment length) ,
+ 𝑑𝑡 𝑚𝑡 ) sin 𝜃1 + 𝑙𝑡 (𝑙𝑠 𝑚𝑓 + 𝑑𝑠 𝑚𝑠 ) 𝜃22̇ sin (𝜃1 + 𝜃2 ) where 𝑚, 𝑐, and 𝐼 are the mass, the center of mass position,
and the mass moment of inertia. The coefficients, 𝑃, 𝑅, and
+ 𝑑𝑓 𝑙𝑡 𝑚𝑓 𝜃32̇ sin (𝜃1 − 𝜃3 ) − 𝑙𝑠 𝑙𝑡 𝑚𝑓 𝜃2̈ cos (𝜃1 + 𝜃2 ) 𝐾, of each segment are listed in Table 2 [60, 61].
6 Journal of Sensors

Table 1: Detailed procedures of inverse dynamics-based static optimization.

Procedures Musculoskeletal model Diagram

Joint
kinematics

Measurement of movements (i) Joint-segment model

Ground
reaction
force

Joint
moments
(i) Body segment parameters (mass, length center
Inverse dynamics
of mass moment of inertia)

Individual
muscle
(i) Muscle groups forces
Static optimization (ii) Moment arms
(iii) Physiological cross sectional area (PCSA)

Table 2: Coefficients for the body segment parameters.

𝑃 𝑅𝑝 𝑅𝑑 𝐾CG 𝐾𝑝 𝐾𝑑
Foot 0.0145 0.500 0.500 0.475 0.690 0.690
Leg 0.0465 0.433 0.567 0.302 0.528 0.643
Thigh 0.1000 0.433 0.567 0.323 0.540 0.653
The subscript means the origin point where the resultant parameter is referring to; 𝑑 refers to the distal joint, 𝑝 refers to the proximal joint, and CG refers to
the center of gravity [60, 61].

2.2.2. Musculoskeletal Model and Static Optimization. Indi- independent: (1) the iliacus (IL) for the hip flexor, (2)
vidual muscular forces were determined by the static the hamstrings (HA) for the hip extensor and knee flexor
optimization based on the moment equilibrium equations simultaneously, (3) the rectus femoris (RF) for the hip
between net joint moment produced by the muscles and flexor and knee extensor, (4) the gastrocnemius (GA) for
the joint moments calculated from the inverse dynamics. the knee flexor and plantar flexor, (5) the biceps femoris
Because the number of muscles is usually larger than that of short head (BF) for the knee flexor, (6) the vastus (VA)
equilibrium equations, an optimization process is required. for the knee extensor, (7) the tibialis anterior (TA) for the
The detailed discussion on applying the static optimization dorsiflexor, (8) the soleus (SO) for the plantar flexor, and (9)
is described in this section, including selecting the muscles the gluteus maximus (GL) for the hip extensor. The muscles
of interest, calculating moments’ arms of each muscle, and have significant force capabilities and moment arms, such
choosing an optimization criterion. The optimization prob- that their contributions to the joint torque are also more
lem was solved by a sequential quadratic programming (SQP) significant than the excluded muscles [62, 63].
algorithm. Even though a muscle exerts only a tensile force, it
As shown in Figure 6, the musculoskeletal model produces torsional force with respect to the joint due to its
included all of the muscle groups which are functionally attachment to the bones. This means that how the muscle
Journal of Sensors 7

y
IL
ẍ p
MH +M GL
dt
RF
(0, 0) ÿ p x

HA VA
mt ; (xt , yt )

휃1 BF

GA
MK

SO TA
ms ; (xs , ys )

휃2 ds

Figure 6: Muscle groups in the sagittal plane (IL: iliacus, HA:

hamstrings, RF: rectus femoris, GA: gastrocnemius, BF biceps
df
femoris short head, VA: vastus, TA: tibialis anterior, SO: soleus, and
GL: gluteus maximus).

MA
m f ; (x f , yf )
tr
o
FGRF

휃3 ham
strin
g s (H

r
tr
A)

Figure 5: Two-dimensional free body diagram of the lower limb

(𝑀𝑖 : joint moment (𝑖 = 𝐻: hip, 𝐾: knee, and 𝐴: ankle), 𝜃𝑗 : absolute
angle of the joints (𝑗 = 1: thigh, 2: shank, and 3: foot), (𝑥𝑘 , 𝑦𝑘 ): center
of mass position of each segment (𝑘 = 𝑡: thigh, 𝑠: shank, and 𝑓: foot),
𝐹GRF : ground reaction, and 𝑟: center of pressure of 𝐹GRF ). rHA

i t = T (
휃k , lt ) T (
휃a , ls ) i s

휃k
generates torque about a joint depends on both muscle HA tr
= i − o tr tr

tendon force and the musculoskeletal geometry. Moment

arms of the muscle groups were calculated using the vector 1 o tr × HA tr
rHA = t
operations based on the coordinate data of the muscle origin 2 儨儨儨儨儨儨儨儨HA tr 儨儨儨儨儨儨儨儨
HA tr
and insertion with respect to a corresponding reference frame i t
[64, 65]. For example, Figure 7 illustrates the vector operation
to calculate the moment arm of the hamstring, where V𝑜tr and
V𝑖𝑠 are the origin and insertion coordinates with respect to
i s
reference coordinate, tr is trunk, and 𝑠 is shank, respectively,
given in [65]. Using the same strategy, moment arms of the
휃a
muscles arms were expressed as a function of joint angles.
To solve the individual muscular forces without reducing
s
the number of muscle groups, an optimization process is
required. It is important to adopt the proper optimization
criteria, and various types have been suggested to simulate the Figure 7: Vector operation for calculating moment arm of the
actual body functions for recruiting the muscles: minimizing hamstring; the subscript identifies the vector (𝑜, 𝑖: origin and
(1) muscle force [18], (2) joint moment [14], (3) ligament insertion), the superscript indicates which reference frame the
force, (4) joint contact force, (5) instantaneous muscle power vector is referring to (tr: trunk, 𝑡: thigh, and 𝑠: shank), 𝜃: is rotational
[20], and (6) muscle stress [15]. Among them, the criterion angle, 𝑙 is segment length, and 𝑇 is homogeneous transformation
of maximum endurance is the most widely accepted recently, matrix as a function of 𝜃 and 𝑙.
8 Journal of Sensors

because it was demonstrated to be the most biologically

meaningful. The cost function to be minimized is as follows:

𝑓𝑖 𝑛 60
𝐽 = √𝑛 ∑ ( ) , (8)
𝐴𝑖
40
where 𝑓𝑖 is the muscular force and 𝐴 𝑖 is the physiological
cross-sectional area (PCSA) of the muscle. The PCSA is

Hip (deg)
defined as the total area of the cross-section perpendicular 20
with respect to the muscle fiber [66], and it is also calculated
as the volume of the muscle divided by the optimal fiber
0
length [67]. An appropriate value of the power 𝑛 in (8) cannot
be determined exactly due to differences in individuals and
muscles. In addition, to specify the value, it is necessary to −20
measure the muscular forces accurately, which would not be 0 20 40 60 80 100
available. Accordingly, the average value in literature reports, Gait cycle (%)
𝑛 = 3, has been adopted as the most appropriate value [16].
80
The optimization problem was specified to minimize the
cost function in (8) satisfying the following constraints:
60
0 ≤ 𝑓𝑖 ≤ (𝑓𝑖 )max , (9)
Knee (deg)
𝐴𝑋 = 𝐵, (10) 40

where 𝑓𝑖 is the exerted muscular force and (𝑓𝑖 )max is the maxi-
mum isometric force. The first constraint in (9) represents the 20
force capability of each muscle, and the maximum isometric
force (𝑓𝑖 )max is determined by multiplying the PCSA and
the specific tension, 61 N/cm2 , of the muscle fiber, which is 0
0 20 40 60 80 100
the maximum stress the fiber can endure. The values of the
Gait cycle (%)
PCSA in the [67] were used. The other constraint in (10)
is moment equilibrium between the joint moment from the 30
inverse dynamics and the resulting moment exerted by the
muscles. The detailed representation of (10) is specified in 20
Appendix.
Ankle (deg)

10
3. Results and Discussion
0
3.1. Verification of the Wearable Sensor System. To verify the
proposed wearable sensor system, joint angles and ground
reactions were measured during normal walking by the −10
proposed sensor system and camera-based motion capture
system (Motion Analysis Corp.) with a force plate (Kistler) −20
0 20 40 60 80 100
simultaneously [68, 69]. A subject (male, 178 cm, 70 kg) with
no musculoskeletal disease walked at a speed of 3 km/h. As Gait cycle (%)
shown in Figure 8, the joint angles were successfully mea- Wearable sensor
sured by the proposed sensor system. In addition, preparation Motion capture
time was less than a minute for attachment of the IMUs and
Figure 8: Joint angles during normal gait.
initialization of joint angles, which was much faster than the
camera-based motion capture system. Ground reactions and
its center of pressure were also successfully measured by the
proposed sensor system (Figure 9). the experimental results were organized according to the four
important functions of a gait cycle (Figure 10); each phase of
3.2. Experimental Results during Walking. Based on the gait cycle was identified by the ground reaction patterns in
motion data shown in Figures 8 and 9, the joint moments Figure 9.
were estimated by solving the inverse dynamics, and the Heel rocker occurs during the phase of initial contact
solution of the optimization problem produced instanta- and the loading response phase, which is required to bear
neous values of muscular forces in the nine muscle groups the large flexion torque of the hip joint due to a significant
at each instant. To interpret the estimated muscular efforts, vertical vector (60% body weight) that is anterior to the hip
Journal of Sensors 9

Forefoot rocker takes place during the terminal stance

(Figure 10(c)). The heel rolled over the anchor of the forefoot
with the ankle virtually locked by the SO and GA, continuing
900 advancement of the shank. Maximal plantar flexion torque
800 was required because the vertical ground reactions, combined
700 with falling body weight, were applied at the toe. The heel rise
posture demanded strong SO and GA action, approximately
600
three times more than in midstance. As the knee began to flex
Force (N)

500 by the end of the terminal stance, the BF muscle was involved
400 in this phase. While the hip was fully extended, the body rolls
300
forward over the forefoot rocker, and the IL and RF muscles
were activated as hip flexors to restrain the rate and extent of
200 hip extension.
100 The limb accomplishes foot clearance in the phase of the
0 initial swing and the midswing (Figure 10(d)). As the shank
0 20 40 60 80 100 became vertical, the ankle joint supported the downward
Gait cycle (%) force of the foot weight. Thus, the TA muscle was activated at
that instant. Because momentum produced by the hip flexion
0.2 was compensated for the pull of gravity on the shank, joint
torque at the knee reached a balance. Additionally, hip flexion
0.15 of the swing phase was generated by the passive gravitational
Center of pressure (m)

force of the limb, such that the intervention of the hip and
0.1 knee muscles was not significant.
The experimental results of individual muscular forces
0.05 during a normal stride are presented in Figure 11(a). To
validate the experimental result, the estimated muscular
0 forces were compared with the another experimental result
(Figure 11(b)); muscular forces were estimated by using a
model-based estimation algorithm, a camera-based motion
−0.05
capture system and a force plate were used to measure
0 20 40 60 80 100 motion information, and EMG signals were also recorded
Gait cycle (%) for validation [63]. The estimated muscular forces show
similar patterns and magnitudes, and the results in this
Insole sensor
study coincide with EMG patterns also. Because it is very
Force plate
difficult to measure true muscular forces of a human, it is not
Figure 9: Ground reactions and center of pressure position with easy to validate the estimated muscular forces quantitatively.
respect to the ankle position during normal gait. However, the activation pattern of muscles during normal
walking was evidenced by well-documented experimental
results with EMG signals [14–16, 19, 20, 54, 70]. Comparing
the estimated muscular forces with the electromyography
joint. Thus, in the estimated result, a large extension torque patterns in the previous studies, it was possible to check
was generated by the hip extensors (HA and GL). Similarly, feasibility of the estimated results.
flexion torque at the knee joint is necessary to endure the
vertical ground reaction force. Due to the two-jointed muscle, 4. Conclusions
HA, a large flexion torque was also produced at the knee joint,
and it was released by the knee extensor, VA. In the case of the Since it is very difficult to measure muscular force directly,
ankle joint, the dorsiflexor TA was excited to endure ground model-based estimation method has been researched actively
reactions placed behind the ankle. In accordance with the for an alternative. Although there exists the commercialized
actions of muscles, the shank made its advance (Figure 10(a)). software package such as OpenSim, the software consists of
Ankle rocker occurs during the midstance to continue a complex musculoskeletal model with higher computation
progression of the body (Figure 10(b)). As the limb rolls costs and technical barrier. Additionally, motions of the
forward over the supporting foot, the directions of the joint human are usually measured camera-based measurement
torques were identically downwards. All of the muscles placed system with a force plate which require time-consuming
on the posterior side of the limb, such as the GL, HA, BF, GA, preparation for attaching reflective markers, data processing
and SO, were involved in the motion. The SO muscle, assisted procedures, and high cost for space and equipment. There-
by the GA, was excited to maintain stance stability of the limb. fore, this study suggests a cost-effective estimation method
Due to the secured stability gained by the action of the SO and by integrating the simple sensor system and the algorithm.
GA muscle, further demand on the quadriceps, including the The proposed sensor system was cost-effective in the aspect of
VA and RF muscle, was minimized. space, equipment expense, preparation, and postprocessing
10 Journal of Sensors

GL
GL

IL
IL

−6∘
∘
25
RF

RF
VA

HA

VA
A
H

BF
BF
10
∘

0∘
GA
GA
SO

SO
TA

TA
0
5∘

0 50 100 0 50 100
Force/max. force (%) Force/max. force (%)
(a) Heel rocker (0%) (b) Ankle rocker (40%)
GL

GL

IL
IL
−12∘

23 ∘

RF
HA

RF

HA

VA
BF
VA

BF

30∘
10 ∘
GA

GA
SO

SO
TA

−5 ∘
TA

∘
13

0 50 100 0 50 100
Force/max. force (%) Force/max. force (%)
(c) Forefoot rocker (60%) (d) Floor clearance (85%)

Figure 10: Visualized experimental results at the four important gait functions. The posture of the segments was based on the measured joint
angles, the green arrows represent directions and relative magnitudes of joint torques estimated by the inverse dynamics, and intensity of the
red color represents the usage of each muscle determined by the static optimization.
Journal of Sensors 11

500

(1) IL

(1) IL
0

1000

(2) HA
(2) HA
500
0

(3) RF
500

(3) RF
0
500 N

(4) GA
(4) GA
1000
0

(5) BF
Force (N)

(5) BF
100
50
0

(6) VA
1000
(6) VA

500

(7) TA
0

500
(7) TA

(8) SO
2000
(8) SO

1000
0

(9) GL
500
(9) GL

0
0 20 40 60 80 100 0 20 40 60 80 100
% gait cycle % gait cycle
(a) (b)

Figure 11: Comparison of the estimated muscular forces: (a) proposed method (b) model-based estimation with camera-based motion capture
system and a force plate.

procedures, and the algorithm involved a simple muscu- Because parameter values from the cadaveric experiment
loskeletal model with lower computation cost. By comparing were adopted instead, a certain level of modeling uncer-
with the camera-based motion capture system with a force tainties would affect the estimated results. In addition, the
plate, the performance of the proposed sensor system was estimated muscular efforts would be feasible under the
verified experimentally, and the estimated muscular forces condition that the human body recruits and activates muscles
were interpreted with respect to the four important gait in the optimal way. It is impossible for the proposed model to
functions and were compared with the previous research with detect abnormal muscle functions such as activation patterns
EMG signals. As a result, it was found that activation patterns of disabled muscles. These are inherent limitation of the
of the muscles were estimated successfully by the proposed inverse dynamics-based static optimization. In this work,
method. the objective was estimating action of the muscular system
Note that there are subject-specific parameters such as to realize the movement measured by the wearable sensor
muscle origin and insertion positions and physiological system rather than finding true value of muscular forces.
cross-sectional areas, which can not be measured directly. However, it would be possible to suggest a special model to
12 Journal of Sensors

estimate muscular forces of the patients with musculoskeletal through the National Research Foundation of Korea (NRF)
disorder such as stroke survivors with additional parameters funded by the Ministry of Science, ICT & Future Planning
or constraints. (NRF-2015R1C1A1A01053763).

Appendix References
The equality constraint in static optimization was specified as [1] V. Zatsiorsky, Kinetics of Human Motion, Human Kinetics, 2002.
follows: [2] S. Jenkins, Sports Science Handbook, Sunningdale Publications,
2001.
𝑓IL [3] R. Teasell, “Stroke recovery and rehabilitation,” Stroke, vol. 34,
[ ]
[𝑓HA ] no. 2, pp. 365-366, 2003.
[ ]
[𝑓 ] [4] H. M. Herr and R. D. Kornbluh, “In New horizons for orthotic
[ RF ] and prosthetic technology: artificial muscle for ambulation,”
[ ]
[𝑓 ] in Proceedings of Smart Structures and Materials 2004: Elec-
[ GA ]
[ ] troactive Polymer Actuators and Devices (EAPAD), vol. 5385 of
𝑋=[ ]
[ 𝑓BF ] , Proceedings of SPIE, July 2004.
[ ]
[ 𝑓VA ] [5] A. T. Asbeck, R. J. Dyer, A. F. Larusson, and C. J. Walsh,
[ ]
[ ] “Biologically-inspired soft exosuit,” in Proceedings of the IEEE
[ 𝑓TA ]
[ ] 13th International Conference on Rehabilitation Robotics (ICORR
[𝑓 ] ’13), pp. 1–8, June 2013.
[ SO ]
[6] Y.-L. Park, B.-R. Chen, D. Young et al., “Bio-inspired active soft
[ 𝑓GL ] orthotic device for ankle foot pathologies,” in Proceedings of
𝑇 the IEEE/RSJ International Conference on Intelligent Robots and
0 0 𝑟IL (A.1) Systems (IROS ’11), pp. 4488–4495, September 2011.
[ ]
[ 0 −𝑟HA,𝑘 −𝑟HA,ℎ ] [7] M. Wehner, B. Quinlivan, P. M. Aubin et al., “A lightweight
[ ] soft exosuit for gait assistance,” in Proceedings of the IEEE
[ 0 𝑟RF,𝑘 𝑟RF,ℎ ]
[ ] International Conference on Robotics and Automation (ICRA
[ ]
[−𝑟 −𝑟 0 ] ’13), pp. 3362–3369, May 2013.
[ GA,𝑎 GA,𝑘 ]
[ ] [8] V. Wank, R. Bauer, B. Walter et al., “Accelerated contrac-
𝐴=[
[ 0 −𝑟 BF 0 ] ,
] tile function and improved fatigue resistance of calf mus-
[ ]
[ 0 𝑟 0 ] cles in newborn piglets with IUGR,” American Journal of
[ VA ] Physiology—Regulatory, vol. 278, no. 2, pp. R304-R310, 2000.
[ ]
[ 𝑟TA 0 0 ]
[ ] [9] C. J. De Luca, “The use of surface electromyography in biome-
[ −𝑟 0 0 ] chanics,” Journal of Applied Biomechanics, vol. 13, no. 2, pp. 135–
[ SO ]
163, 1997.
[ 0 0 −𝑟GL ] [10] N. J. Cronin and G. Lichtwark, “The use of ultrasound to study
𝑇 muscle-tendon function in human posture and locomotion,”
𝐵 = [𝑀𝑎 𝑀𝑘 𝑀ℎ ] , Gait and Posture, vol. 37, no. 3, pp. 305–312, 2013.
[11] O. Amft, H. Junker, P. Lukowicz, G. Tröster, and C. Schuster,
where 𝑋 refers to a vector of muscular forces, 𝑟 means the “Sensing muscle activities with body-worn sensors,” in Proceed-
moment arm of the muscle with respect to the corresponding ings of the International Workshop on Wearable and Implantable
joint in the subscript, 𝑎, 𝑘, and ℎ mean the ankle, knee, and Body Sensor Networks (BSN ’06), pp. 138–141, April 2006.
hip joints, respectively, and 𝑀 is the joint moment calculated [12] S. L. Delp, F. C. Anderson, A. S. Arnold et al., “OpenSim: open-
by the inverse dynamics. source software to create and analyze dynamic simulations of
movement,” IEEE Transactions on Biomedical Engineering, vol.
54, no. 11, pp. 1940–1950, 2007.
Disclosure
[13] D. G. Thelen, F. C. Anderson, and S. L. Delp, “Generating
Preliminary versions of this paper were presented at the dynamic simulations of movement using computed muscle
IEEE/ASME International Conference on Advanced Intelli- control,” Journal of Biomechanics, vol. 36, no. 3, pp. 321–328,
gent Mechatronics (AIM) 2015. 2003.
[14] J. J. Collins, “The redundant nature of locomotor optimization
laws,” Journal of Biomechanics, vol. 28, no. 3, pp. 251–267, 1995.
Conflicts of Interest [15] R. D. Crowninshield, R. C. Johnston, J. G. Andrews, and R.
A. Brand, “A biomechanical investigation of the human hip,”
The authors declare that they have no conflicts of interest. Journal of Biomechanics, vol. 11, no. 1-2, pp. 75–85, 1978.
[16] R. D. Crowninshield and R. A. Brand, “A physiologically based
Acknowledgments criterion of muscle force prediction in locomotion,” Journal of
Biomechanics, vol. 14, no. 11, pp. 793–801, 1981.
This work was supported by the 2016 Research Fund [17] U. Glitsch and W. Baumann, “The three-dimensional deter-
(1.160005.01) of UNIST (Ulsan National Institute of Sci- mination of internal loads in the lower extremity,” Journal of
ence and Technology) and Basic Science Research Program Biomechanics, vol. 30, no. 11-12, pp. 1123–1131, 1997.
Journal of Sensors 13

[18] A. G. Patriarco, R. W. Mann, S. R. Simon, and J. M. Mansour, contact forces and muscle forces evaluated in wheelchair-
“An evaluation of the approaches of optimization models in related activities of daily living in able-bodied subjects versus
the prediction of muscle forces during human gait,” Journal of subjects with paraplegia and tetraplegia,” Archives of Physical
Biomechanics, vol. 14, no. 8, pp. 513–525, 1981. Medicine and Rehabilitation, vol. 86, no. 7, pp. 1434–1440, 2005.
[19] D. R. Pedersen, R. A. Brand, and D. T. Davy, “Pelvic muscle and [35] D. Vlasic, R. Adelsberger, G. Vannucci et al., In Practical
acetabular contact forces during gait,” Journal of Biomechanics, Motion Capture in Everyday Surroundings, ACM Transactions
vol. 30, no. 9, pp. 959–965, 1997. on Graphics (TOG), 2007.
[20] H. Röhrle, R. Scholten, C. Sigolotto, W. Sollbach, and H. Kellner, [36] T. Cloete and C. Scheffer, “Benchmarking of a full-body inertial
“Joint forces in the human pelvis-leg skeleton during walking,” motion capture system for clinical gait analysis,” in Proceedings
Journal of Biomechanics, vol. 17, no. 6, pp. 409–424, 1984. of the 30th Annual International Conference of the IEEE Engi-
[21] A. Seireg and R. J. Arvikar, “The prediction of muscular neering in Medicine and Biology Society (EMBS ’08), pp. 4579–
load sharing and joint forces in the lower extremities during 4582, Vancouver, Canada, August 2008.
walking,” Journal of Biomechanics, vol. 8, no. 2, pp. 89–102, 1975. [37] M. Brodie, A. Walmsley, and W. Page, “Fusion motion capture: a
[22] A. Schultz, K. Haderspeck, D. Warwick, and D. Portillo, “Use prototype system using inertial measurement units and gps for
of lumbar trunk muscles in isometric performance of mechan- the biomechanical analysis of ski racing,” Sports Technology, vol.
ically complex standing tasks,” Journal of Orthopaedic Research, 1, pp. 17–28, 2008.
vol. 1, no. 1, pp. 77–91, 1983. [38] D. Roetenberg, H. Luinge, and P. Slycke, Xsens mvn: Full 6dof
[23] J. C. Bean, D. B. Chaffin, and A. B. Schultz, “Biomechanical Human Motion Tracking Using Miniature Inertial Sensors, Xsens
model calculation of muscle contraction forces: a double linear Motion Technologies BV, Tech., 2009.
programming method,” Journal of Biomechanics, vol. 21, no. 1, [39] Y. Zheng, K.-C. Chan, and C. C. L. Wang, “Pedalvatar: an IMU-
pp. 59–66, 1988. based real-time body motion capture system using foot rooted
[24] R. E. Hughes, D. B. Chaffin, S. A. Lavender, and G. B. J. kinematic model,” in Proceedings of the IEEE/RSJ International
Andersson, “Evaluation of muscle force prediction models of Conference on Intelligent Robots and Systems (IROS ’14), pp.
the lumbar trunk using surface electromyography,” Journal of 4130–4135, September 2014.
Orthopaedic Research, vol. 12, no. 5, pp. 689–698, 1994. [40] Xsens, “MVN Biomech,” https://www.xsens.com/products/
[25] W. Z. Kong, V. K. Goel, and L. G. Gilbertson, “Prediction of mvn-biomech/.
biomechanical parameters in the lumbar spine during static [41] Perception Neuron, “Perception Neuron,” 2015, https://
sagittal plane lifting,” Journal of Biomechanical Engineering, vol. neuronmocap.com/products/perception_neuron.
120, no. 2, pp. 273–280, 1998. [42] Synertial, “IGS-Cobra Professional Mo-Cap Solutions,” 2015,
[26] M. A. Nussbaum and D. B. Chaffin, “Lumbar muscle force http://synertial.com/products/.
estimation using a subject-invariant 5-parameter EMG-based [43] Meta motion, “IGS 190 Motion Capture System,” 2015, http://
model,” Journal of Biomechanics, vol. 31, no. 7, pp. 667–672, 1998. www.metamotion.com/gypsy/gypsy-gyro.htm.
[27] S. H. M. Brown and J. R. Potvin, “Constraining spine stability [44] Technaid, “Motion Capture System,” 2015, http://www.technaid
levels in an optimization model leads to the prediction of trunk .com/en/products/motion-capture-system.
muscle cocontraction and improved spine compression force [45] G. Cooper, I. Sheret, L. McMillian et al., “Inertial sensor-based
estimates,” Journal of Biomechanics, vol. 38, no. 4, pp. 745–754, knee flexion/extension angle estimation,” Journal of Biomechan-
2005. ics, vol. 42, no. 16, pp. 2678–2685, 2009.
[28] N. Brook, J. Mizrahi, M. Shoham, and J. Dayan, “A biomechan- [46] T. Khurelbaatar, K. Kim, S. Lee, and Y. H. Kim, “Consistent
ical model of index finger dynamics,” Medical Engineering and accuracy in whole-body joint kinetics during gait using wear-
Physics, vol. 17, no. 1, pp. 54–63, 1995. able inertial motion sensors and in-shoe pressure sensors,” Gait
[29] D. D. Penrod, D. T. Davy, and D. P. Singh, “An optimization and Posture, vol. 42, no. 1, pp. 65–69, 2015.
approach to tendon force analysis,” Journal of Biomechanics, vol. [47] J.-C. Kim, K.-S. Kim, and S. Kim, “Wearable sensor system
7, no. 2, pp. 123–129, 1974. including optical 3-axis GRF sensor for joint torque estimation
[30] R. Raikova and H. Aladjov, “The influence of the way the muscle in real-time gait analysis,” in Proceedings of the IEEE/ASME
force is modeled on the predicted results obtained by solving International Conference on Advanced Intelligent Mechatronics
indeterminate problems for a fast elbow flexion,” Computer (AIM ’14), pp. 112–117, July 2014.
Methods in Biomechanics and Biomedical Engineering, vol. 6, no. [48] S. J. M. Bamberg, A. Y. Benbasat, D. M. Scarborough, D. E.
3, pp. 181–196, 2003. Krebs, and J. A. Paradiso, “Gait analysis using a shoe-integrated
[31] J. Pierce and G. Li, “Muscle forces predicted using optimization wireless sensor system,” IEEE Transactions on Information
methods are coordinate system dependent,” Journal of Biome- Technology in Biomedicine, vol. 12, no. 4, pp. 413–423, 2008.
chanics, vol. 38, pp. 695–702, 2005. [49] N. K. Lee, R. S. Goonetilleke, Y. S. Cheung, and G. M. So,
[32] R. E. Hughes, M. G. Rock, and K.-N. An, “Identification of “A flexible encapsulated MEMS pressure sensor system for
optimal strategies for increasing whole arm strength using biomechanical applications,” Microsystem Technologies, vol. 7,
Karush-Kuhn-Tucker multipliers,” Clinical Biomechanics, vol. no. 2, pp. 55–62, 2001.
14, no. 9, pp. 628–634, 1999. [50] S. Kim, K. Ro, and J. Bae, “Real-time estimation of individual
[33] H.-T. Lin, F.-C. Su, H.-W. Wu, and K.-N. An, “Muscle forces muscular forces of the lower limb using wearable sensors,”
analysis in the shoulder mechanism during wheelchair propul- in Proceedings of the IEEE/ASME International Conference
sion,” Journal of Engineering in Medicine, vol. 218, pp. 213–221, on Advanced Intelligent Mechatronics (AIM ’15), pp. 432–436,
2004. Pusan, South Korea, July 2015.
[34] S. Van Drongelen, L. H. Van Der Woude, T. W. Janssen, E. L. [51] Withrobot, “myahrs+,” 2017, http://www.withrobot.com/
Angenot, E. K. Chadwick, and D. H. Veeger, “Glenohumeral myahrs_en/.
14 Journal of Sensors

[52] J. Bae, K. Kong, N. Byl, and M. Tomizuka, “A mobile gait

monitoring system for abnormal gait diagnosis and rehabili-
tation: a pilot study for Parkinson disease patients,” Journal of
Biomechanical Engineering, vol. 133, no. 4, p. 041005, 2011.
[53] J. Bae and M. Tomizuka, “A tele-monitoring system for gait
rehabilitation with an inertial measurement unit and a shoe-
type ground reaction force sensor,” Mechatronics, vol. 23, no. 6,
pp. 646–651, 2013.
[54] J. Perry and J. Davids, “Gait analysis: normal and pathological
function,” Journal of Pediatric Orthopaedics, vol. 12, p. 815, 1992.
[55] G. Li, T. Liu, L. Gu, Y. Inoue, H. Ning, and M. Han, “Wearable
gait analysis system for ambulatory measurement of kinematics
and kinetics,” in Proceedings of the 13th IEEE SENSORS Confer-
ence (SENSORS ’14), pp. 1316–1319, November 2014.
[56] J. Bae, K. Kong, and M. Tomizuka, “Real-time estimation of
lower extremity joint torques in normal gait,” Robot Control, vol.
9, pp. 443–448, 2009.
[57] K. Kong and M. Tomizuka, “A gait monitoring system based
on air pressure sensors embedded in a shoe,” IEEE/ASME
Transactions on Mechatronics, vol. 14, no. 3, pp. 358–370, 2009.
[58] B. Bresler and J. Frankel, “The forces and moments in the leg
during level walking,” Transactions of the American Society of
Mechanical Engineers, vol. 72, pp. 25–35, 1950.
[59] D. A. Winter, Biomechanics and Motor Control of Human
Movement, John Wiley & Sons, 2009.
[60] D. Miller and R. Nelson, Biomechanics of Sports, Lea and
Febiger, 1973.
[61] Dempster, Patterns of Human Motion, Prentice-Hall, 1971.
[62] S. Heintz, “Muscular forces from static optimization,” 2006.
[63] A. Pedotti, V. V. Krishnan, and L. Stark, “Optimization of
muscle-force sequencing in human locomotion,” Mathematical
Biosciences, vol. 38, no. 1-2, pp. 57–76, 1978.
[64] M. G. Hoy, F. E. Zajac, and M. E. Gordon, “A musculoskeletal
model of the human lower extremity: the effect of muscle,
tendon, and moment arm on the moment-angle relationship of
musculotendon actuators at the hip, knee, and ankle,” Journal of
Biomechanics, vol. 23, no. 2, pp. 157–169, 1990.
[65] R. A. Brand, R. D. Crowninshield, C. E. Wittstock, D. R.
Pedersen, C. R. Clark, and F. M. Van Krieken, “Model of
lower extremity muscular anatomy,” Journal of Biomechanical
Engineering, vol. 104, pp. 304–310, 1982.
[66] R. J. Maughan, J. S. Watson, and J. Weir, “Strength and
cross-sectional area of human skeletal muscle,” The Journal of
Physiology, vol. 338, no. 1, pp. 37–49, 1983.
[67] E. M. Arnold, S. R. Ward, R. L. Lieber, and S. L. Delp, “A model
of the lower limb for analysis of human movement,” Annals of
Biomedical Engineering, vol. 38, no. 2, pp. 269–279, 2010.
[68] Motion Analysis Corp, “Kestrel digital realtime system,” 2017,
http://www.motionanalysis.com/html/movement/kestrel.html.
[69] Kistler, “Multicomponent force plate,” 2017, https://www.kistler
.com/?type=669&fid=60693&.
[70] A. Cappozzo, T. Leo, and A. Pedotti, “A general computing
method for the analysis of human locomotion,” Journal of
Biomechanics, vol. 8, no. 5, pp. 307–320, 1975.
 International Journal of

Rotating
Machinery

International Journal of
The Scientific
(QJLQHHULQJ Distributed
Journal of
Journal of

Hindawi Publishing Corporation

World Journal
Hindawi Publishing Corporation Hindawi Publishing Corporation
Sensors
Hindawi Publishing Corporation
Sensor Networks
Hindawi Publishing Corporation
http://www.hindawi.com Volume 201 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

https://www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi Publishing Corporation
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

VLSI Design
Advances in
OptoElectronics
,QWHUQDWLRQDO-RXUQDORI

International Journal of
Modelling &
Simulation
$HURVSDFH
Hindawi Publishing Corporation Volume 2014
Navigation and
Observation
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
in Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
(QJLQHHULQJ
+LQGDZL3XEOLVKLQJ&RUSRUDWLRQ
KWWSZZZKLQGDZLFRP 9ROXPH
Hindawi Publishing Corporation
http://www.hindawi.com
http://www.hindawi.com Volume 201-

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014