Kinect V2 for Upper Limb Rehabilitation Applications

A Preliminary Analysis on Performance Evaluation

Giorgia Lupinacci, Gianluca Gatti, Agostino Angilica and Maurizio Muzzupappa

Dept. of Mechanical, Energy and Management Engineering (DIMEG), University of Calabria,

V. P. Bucci, Arcavacata di Rende (CS), 87036, Italy

Keywords: Upper Limb Rehabilitation, Upper Limb Joints Tracking, Kinect V2.

Abstract: Many systems have been developed to facilitate upper limb rehabilitation procedures in human subjects

affected by trauma or pathologies and to retrieve information about patient performance. The Microsoft

Kinect sensor can be used in this context to track body motion and detect objects. In order to evaluate the

usability of this device in the upper limb rehabilitation field, a comparison with a marker-based system is

presented in this paper. The upper limb motion is specifically considered and the performance on its

detection and tracking is evaluated. The effect of the relative location between the Kinect and the observed

subject is also investigated through experimental tests performed in different configurations.

1 INTRODUCTION

Modern upper limb rehabilitation systems use

motion tracking technologies to evaluate patient

performance, to aid limb motion in robotic assisted

rehabilitation techniques or to allow interaction with

virtual reality (Zhou and Hu, 2008). Shoulder, elbow

and wrist joints tracking, is a necessary step for

motion classification and recognition (Pimentel do

Rosàrio, 2014). Furthermore, upper limb tracking

allows to develop patient-specific therapies and is

used to create serious games that induce motor

recovery in a stimulating environment (Burke et al.,

2009). Assisted motion control is also fundamental

for the application of autonomous or semi-

autonomous systems for motor recovery and to make

rehabilitative experience possible at home (Prieto et

al.,

2014). Marker-less motion tracking systems for

gaming are proposed as an alternative to marker-

based systems (Moeslund et al., 2006; Lange et al.,

2011) (usually adopted for motion analysis), because

they are easy to use and less expensive. Moreover, a

marker-less system is more flexible, because it

offers the advantage to be used in several

environments and configurations.

One of the most interesting marker-less systems

available nowadays is the Microsoft Kinect

(Kinect for Windows features, 2015). The device is

able to retrieve the position of 25 human body joints

and track up to six human subjects at the same time.

It predicts the body joints from a single depth image.

From July 2014 a new version of the sensor is

available, namely the Kinect for Windows v2.0, with

improved performances respect to the first release,

thanks to the higher fidelity of depth images (Lachat

et al., 2015).

Kinect performance was studied using three main

approaches: posture detection (Clark et al., 2012;

van Diest et al., 2014; Diego-Mas and Alcaide-

Marzal, 2014), joints centre evaluation (Xu and

McGorry, 2015) and angles evaluation (Bonnechère

et al., 2014; Chang et al., 2012). In particular, Xu

and McGorry (Xu and McGorry, 2015) very recently

proposed a comparison between the first and second

generation of the Kinect and a marker-based system,

concluding that no impressive improvements are

introduced by the new version of the Kinect if the

tracking is assessed at the whole skeleton level for

static posture evaluation.

It is to be noted, however, that most of the

available studies consider only the first generation of

the Kinect (Bonnechère et al., 2014; Chang et al.,

2012; Clark et al., 2012; van Diest et al., 2014), so

the advantages of the second generation are not

completely addressed, especially in the upper limb

rehabilitation field.

The aim of this work is thus to validate the

reliability of the second generation of the Kinect

when adopted for upper limb rehabilitation. The

study was conducted by taking into consideration (i)

Lupinacci, G., Gatti, G., Angilica, A. and Muzzupappa, M.

Kinect V2 for Upper Limb Rehabilitation Applications - A Preliminary Analysis on Performance Evaluation.

DOI: 10.5220/0005659201310138

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 1: BIODEVICES, pages 131-138

ISBN: 978-989-758-170-0

131

the possibility to use the sensor in different positions

without calibration and (ii) the body tracking

algorithm, already integrated in the Kinect software.

A marker-based system was used to conduct the

evaluation of the upper limb tracking performance.

The upper limb motion was executed by an healthy

subject guided by the use of a specific end-effector

as to emulate the movements usually asked to the

patient during a robot-assisted rehabilitative session,

e.g. the “reaching” and the “side to side” exercise

(Volpe et al., 2008; Frisoli et al., 2012; Lam et al.,

2008). The performances of the Kinect are compared

to that of a marker-based system, for specific

complex movements, affecting both the shoulder and

elbow rotation simultaneously.

2 MATERIALS AND METHODS

Upper limb tracking was performed using the second

version of the Microsoft Kinect for Windows and a

stereo-photogrammetric optical system, the

Optitrack (NaturalPoint, Inc. - Optical Tracking

Solutions, 2015), consisting of 8 Flex13 cameras

acquiring up to a frequency of 120 Fps. For body

tracking, the Kinect body model and joints position

were obtained from the associated Software

Development Kit (SDK) v2.0 (Microsoft, Kinect for

Windows, 2015), working at up to 30 Fps. For the

Optitrack system, the NatNet SDK v2.7 (Natural

Point Inc. - Optical Tracking Solutions, 2015) and

the Motive 1.7.5 software (Natural Point Inc. -

Optical Tracking Solutions, 2015) was adopted to

capture the body joints position in conjunction with

a motion capture suit and reflective markers. Figure

1 shows the set up for the upper limb tracking. In the

figures both the Kinect sensor and the Optitrack

cameras are indicated, as well as the corresponding

coordinate systems used to retrieve numerical

values.

Figure 1: Experimental setup for upper limb tracking.

2.1 Upper Limb Tracking

During the study presented in this work, it was asked

an healthy subject to make a set of movements using

a passive end-effector. In particular, the end-effector

which was visible in Figure 1 and is now shown in

the detail of Figure 2, consisted of a tripod and a bar

fixed on it. It was made to constrain the hand

movement of the subject to a circular or a linear

motion, so that a reference trajectory was available

for validation. Each session was recorded by using

both the Kinect and the Optitrack system.

Figure 2: End-effector used for upper limb motion. It

consists of a tripod and a bar.

Figure 3: Schematic representation of Kinect position

relative to the subject for upper limb tracking. Parameters

are given in Table 1.

The healthy subject was seated on a stool, placed

in the field of view of the Optitrack cameras and the

Kinect was placed in six different positions as

illustrated in Figure 3, but oriented so as it was

facing the stool. The different positions where the

Kinect was located are defined to be at the right,

middle and left position relative to the axis

orthogonal to the subject front plane, as defined in

Table 1. For each of these positions, the Kinect was

located at two different distances from the floor,

respectively at about 800 mm and 1500 mm. In

Table 1 these positions are summarized and labeled

from K1 to K6 for ease of reference. The subject

was wearing a motion capture suit with reflective

markers to identify the shoulder, elbow and wrist

joint, as shown in Figure 1. Such analysis is

BIODEVICES 2016 - 9th International Conference on Biomedical Electronics and Devices

132

performed to investigate if the Kinect position

relative to the subject was somehow affecting the

overall tracking procedure. A circular motion and a

linear motion were thus executed by the right hand

of the healthy subject for each position of the

Kinect, as described above. It is worth noting that

the healthy subject was asked not to rotate the wrist

during the constrained motion, even if some small

effect due to this compliant requirement are

expected to affect the obtained results reported in

Section 3.

Figure 4: Schematic model of the upper limb with

geometric parameters and joints variable, as defined in

(Mihelj, 2006).

Table 1: Definition of the six Kinect position used for

upper limb tracking.

Position a [mm]

b[mm] c [mm]

K1 – right down 800 800 -2000

K2 – right up 800 1500 -2000

K3 – middle down 0 800 -2000

K4 – middle up 0 1500 -2000

K5 – left down -800 800 -2000

K6 – left up -800 1500 -2000

The Kinect and the Optitrack frames were

synchronized using the timestamp for direct

comparison.

To determine the shoulder and elbow angles

from the coordinates of the position of the shoulder,

elbow and wrist joints obtained from the two optical

systems, a 4 degree-of-freedom (DOF) inverse

kinematic model of the human arm was used

(Mihelj, 2006). A schematic model of the upper limb

with geometric parameters is represented in Figure

4. The shoulder is modeled as a ball-and-socket joint

with rotation axis for abduction-adduction (angle

), flexion-extension (angle q

) and internal-

external rotation (angle q

With reference to Figure 4, l

and l

are,

respectively, the arm and forearm lengths. The

RMSDs between Optitrack angles variation and

Kinect angles variation were computed, to give an

indication of the difference between the two

systems.

The trajectory described by the wrist as

computed from the two optical system for each

session, is also used to assess the performance of the

two motion tracking systems. In particular, a best-fit

plane was determined on the set of points

corresponding to the wrist position during motion.

This is because the motion constrained by the end-

effector was planar and this allowed to preliminary

filter the results. Once projected on the

corresponding best-fit plane, a second fitting was

performed on the projected trajectories to fit a circle

and a line, respectively, for the circular motion and

for the linear motion of the hand grasping the end-

effector. The RMSD of the residual errors of each

fitting procedure, the radius of the fitted circle and

the range of motion (RoM) for each trajectory were

computed.

3 RESULTS

For upper limb tracking, the shoulder and elbow

angles are computed for each of the 12 sessions (one

circular and one translational motion for the six

Kinect configurations). The sample in Figure 5

represents the results relative to the circular motion

executed by the participant using the end-effector,

with the Kinect at the middle down position K3, as

defined in Table 1. In Figure 5(a-d) the variation of

the angles q

,…q

, are respectively showed, as

obtained from the Kinect and Optitrack joints data,

for a circular motion of the end-effector with the

Kinect in position K3. Also, in Figure 5(e) the

trajectories computed using the Kinect and the

Optitrack data are showed.

Figure 6 illustrates the same information

represented in Figure 5, but relative to a circular

motion captured with the Kinect at the right down

position K1, as defined in Table 1. Figure 7 and

Figure 8 represent the results relative to the linear

motion with the Kinect at the middle down position

K3 and at the right down position K1, respectively.

It is worth noting that the parameters of the

Kinect and Optitrack body model do not correspond

to each other, and this is because of the intrinsic

identification of joints location performed by the two

systems. This means, in particular, that the arm and

forearm length, as identified by the systems, are in

general different, and they also vary along the

acquisition time. For the Kinect, the identified joints

location are affected mainly by the intrinsic system

Shoulder

Elbow

Arm

Kinect V2 for Upper Limb Rehabilitation Applications - A Preliminary Analysis on Performance Evaluation

133

setting, ambient light and sensor resolution. For the

Optitrack, they are affected mainly by the body suit

and the reflective marker location. To have a better

insight into these effects, Table 2 and 3 reports the

mean and STD of l

and l

as identified by the two

optical systems, for the circular and linear motion of

the end-effector, respectively.

Comparing Figure 5 to Figure 6, it can be seen

that a better match between Optitrack and Kinect

data is achieved for the first dataset. Furthermore,

the wrist trajectory computed by the Kinect (Figure

6 (e)) presents some noise due to the high variation

in the identification of the forearm and arm lengths,

as explained above. In particular, from Table 2, it is

observed that the STDs of l

and l

obtained by the

Kinect at the middle down pose (K3), are the lowest

compared to all the Kinect acquisitions. The STD of

when the Kinect is in position K5 is lower than for

K3 but the STD of l

is greater than for K3. The same

observations can be done comparing Figure 7 and

Figure 8. In the middle down configuration, indeed,

the subject was completely visible and no joints

occlusion occurred during the acquisitions.

In Table 4 and Table 5 the RMSDs obtained

from the angles variations of the Optitrack and

Kinect data are reported, for circular motions and

linear motions, respectively. The lower trend of

these values for middle positions (K3 and K4) of the

Kinect confirms that this is the preferable position

for the sensor. Tables 6 and 7 summarize the results

obtained in terms of the wrist trajectory for the

circular and linear motion, respectively. In Table 6

the radius associated to each circular trajectory after

circle fitting, and the RoM are reported, while for

the linear motion the RoM is reported in Table 7.

Table 2: Mean value  of the arm length l

, and forearm

length l

identified by the two optical systems along the

acquisition frames, for the circular motion and for the

different Kinect positions as from Table 1. The value in

parenthesis is the STD () of the correspondent

distribution. The sub-script “K” stands for Kinect, while

the “O” stands for Opitrack.

Circular Motions

K1 K2 K3 K4 K5 K6

[mm]



()

280.6

(12.6)

204.6

(18.0)

202

(4.4)

224.6

(20.5)

188.1

(2.3)

230.7

(10)



()

286.0

(6E-4)

286.0

(7E-4)

276.5

(4E-3)

286

(1E-3)

286

(1E-3)

286

(8E-4)

[mm]



()

234

(3)

213.5

(18.3)

203.6

(0.8)

215.3

(11.2)

192

(11.1)

223

(8.6)



()

285

(7.2)

275.4

(12.0)

252.0

(12.5)

263.2

(9.4)

253.3

(2)

268.5

(9.2)

Table 3: Mean value  of the arm length l

, and forearm

length l

identified by the two optical systems along the

acquisition frames, for the circular motion and for the

different Kinect positions as from Table 1. The value in

parenthesis is the STD () of the correspondent

distribution. The sub-script “K” stands for Kinect, while

the “O” stands for Opitrack.

Linear Motions

K1 K2 K3 K4 K5 K6

[mm]



()

230.5

(11.5)

222.4

(7.7)

248

(4.1)

211

(19.4)

243.5

(6.8)

230.7

(10)



()

286.0

(1E-3)

286

(4E-4)

286

(6E-4)

286

(4E-4)

286

(4E-4)

285.8

(9E-4)

[mm]



()

239.7

(25.3)

216.0

(7)

187.5

(1.6)

232.6

(16.3)

222

(7.2)

223

(8.5)



()

274.0

(4.4)

260.4

(8.4)

236.0

(8.2)

243.6

(7.5)

268.2

(14)

268.5

(9.2)

Table 4: RMSDs obtained comparing the arm angles

variations for Kinect and Optitrack for the circular

motions of the end-effector and for the six different Kinect

configurations from Table 1.

Circular Motions

K1 K2 K3 K4 K5 K6

RMSD q

2.5 32.2 6 1.1 5.7 3

RMSD q

2.0 10.6 3.1 4.5 5.7 4

RMSD q

20.0 53.0 21.2 4.0 2.6 31

RMSD q

4.9 17.8 8.4 11.8 3.6 2.7

Table 5: RMSDs obtained comparing the arm angles

variations for Kinect and Optitrack for the linear motion of

the end-effector and for the six different Kinect

configurations from Table 1.

Linear Motions

K1 K2 K3 K4 K5 K6

RMSD q

17.05 7.4 6.7 7.9 3.8 6.7

RMSD q

16.1 9.2 4.5 1.1 4.5 1.8

RMSD q

27.7 7.6 8.5 6.0 20.4 32.0

RMSD q

16.4 5.7 8.4 4.4 7.7 3.7

Table 6: Results of the wrist trajectories for circular

motion. Circle-fitted radius and estimated RoM, for the six

different Kinect configurations as from Table 1. The “K”

stands for Kinect and the “O” stands for Optitrack.

K1 K2 K3 K4 K5 K6

Radius

[mm]

K 180 102 95.6 118 122 183

O 179.1 163.3 119.6 201.4 198.7 176.5

RoM

[deg]

K 105.2 111.7 129.0 96.5 114.3 148

O 92.3 102.1 121.3 76.8 111.8 173.3

Table 7: Results of the wrist trajectories for linear motion.

Estimated RoM, for the six different Kinect configurations

as from Table 1. The “K” stands for Kinect and the “O”

stands for Optitrack.

K1 K2 K3 K4 K5 K6

RoM

[mm]

K 234.2 240 263.4 228.5 273.3 293.8

O 312.2 326 284.9 308.6 292.2 316.9

BIODEVICES 2016 - 9th International Conference on Biomedical Electronics and Devices

134

(a)

(b)

(c)

(d)

(e)

Figure 5: Results for the circular motion of the end-

effector using the right arm, with Kinect in position K3.

(a)



, (b)



, (c)



, (d)



, (e) estimated wrist

trajectory by Optitrack (spotted lines) and Kinect

(discontinue lines). The axes x and z represent the

coordinates of the best-fit plane determined on the set of

points corresponding to the wrist position during motion.

For the circular motions, the mean value of the

difference between the radius of the circle fitted on

the Kinect data and on the Optitrack data is 42(36)

mm, where the value in parentheses represents the

(a)

(b)

(c)

(d)

(e)

Figure 6: Results for the circular motion of the end-

effector using the right arm, with Kinect in position K2.

(a)



, (b)



, (c)



, (d)



, (e) estimated wrist

trajectory by Optitrack (spotted lines) and Kinect

(discontinue lines). x and z are the coordinate of the best-

fit plane determined on the set of points corresponding to

the wrist position during motion.

STD of the correspondent distribution. Instead, the

mean difference between the ROM computed from

the Kinect data and from the Optitrack data, is

13(8.3) deg. Finally, the mean difference between

0 10 20 30 40 50

-5

Frame



[deg]

0 10 20 30 40 50

-16

-14

-12

-10

-8

-6

-4

-2

Frame



[deg]

0 10 20 30 40 50

Frame



[deg]

0 10 20 30 40 50

-50

-40

-30

-20

-10

Frame



[deg]

-150 -100 -50 0 50 100 150

-100

-50

100

x [mm]

z [mm]

0 20 40 60 80 100

-10

Frame



[deg

]

0 20 40 60 80 100

-70

-60

-50

-40

-30

-20

-10

Frame



[deg]

0 20 40 60 80 100

-60

-40

-20

Frame



[deg]

0 20 40 60 80 100

100

Frame



[deg]

-150 -100 -50 0 50 100 150

-150

-100

-50

100

x [mm]

z [mm]

Kinect V2 for Upper Limb Rehabilitation Applications - A Preliminary Analysis on Performance Evaluation

135

(a)

(b)

(c)

(d)

(e)

Figure 7: Results for the linear motion of the end-effector

using the right arm, with Kinect in position K3. (a)



(b)



, (c)



, (d)



, (e) estimated wrist trajectory by

Optitrack (spotted lines) and Kinect (discontinue lines).

The axes x and z represent the coordinates of the best-fit

plane determined on the set of points corresponding to the

wrist position during motion.

the ROM on the linear motion of the hand on the bar

is equal to 51.4(33.3) mm.

From Figures 5, Figure 6, Figure 7 and Figure 8

(a)

(b)

(c)

(d)

(e)

Figure 8: Results for the linear motion of the end-effector

using the right arm, with Kinect in position K2. (a)



(b)



, (c)



, (d)



, (e) estimated wrist trajectory by

Optitrack (spotted lines) and Kinect (discontinue lines). x

and z are the coordinate of the best-fit plane determined on

the set of points corresponding to the wrist position during

motion.

it is seen that there are some offsets between the

angles estimated by the two systems, and this is due

to the different body model inherently adopted by

0 20 40 60 80 100 120 140

-30

-25

-20

-15

-10

-5

Frame



[deg]

0 20 40 60 80 100 120 140

-5

-4

-3

-2

-1

Frame



[deg]

0 20 40 60 80 100 120 140

-60

-50

-40

-30

-20

-10

Frame



[deg]

0 20 40 60 80 100 120 140

-5

Frame



[deg]

-100 -50 0 50 100 150 200

-15

-10

-5

x [mm]

z [mm]

0 10 20 30 40 50 60 70 80

-25

-20

-15

-10

-5

Frame



[deg]

0 10 20 30 40 50 60 70 80

-8

-6

-4

-2

Frame



[deg]

0 10 20 30 40 50 60 70 80

-45

-40

-35

-30

-25

-20

-15

-10

-5

Frame



]

0 10 20 30 40 50 60 70 80

-5

Frame



[deg]

-150 -100 -50 0 50 100 150 200 250

-15

-10

-5

x [mm]

z [ mm]

BIODEVICES 2016 - 9th International Conference on Biomedical Electronics and Devices

136

the Kinect and Optitrack, as discussed above. For

this reason it is not expected that the angles obtained

from the data captured by the two systems are the

same. The comparison between the two systems in

terms of RoM should be then taken as a qualitative

indication only.

4 DISCUSSION

This paper presented a preliminary investigation on

the expected results and performance that could be

obtained when using the marker-less Kinect sensor

for upper limb motion tracking, with specific

application to the rehabilitation of the upper limb. In

particular, the aim of this study was focused on the

Kinect v2, recently released by the Microsoft. An

auxiliary marker-based Optitrack system, composed

by 8 cameras and motion capture suit with reflective

markers, was used as reference. The Kinect marker-

less system can be used to detect objects in the 3D

space with an accuracy close to the centimetre

(Lachat et al., 2015). The accuracy of the Optitrack

system has been estimated to be of the order of the

millimetre (Carse et al., 2013). So, the Optitrack

system exhibits an accuracy higher than that of the

Kinect, and this confirm its use as a reference.

For the upper limb tracking procedure, a realistic

experimental setup was created. An healthy subject

(participant) was asked to execute specific motions

grasping an end-effector, in order to reproduce a

robotic-assisted rehabilitation procedure (Zhou and

Hu, 2008). It was observed that, in order to have a

better precision and accuracy, the Kinect should be

located in front of the subject. The comparison

between the two optical systems could only give

qualitative indication on the relative precision and

accuracy in detecting upper limb tracking. This was

related to the inherently different body models

implemented in the Kinect and Optitrack software

tools. Furthermore, on the one hand the Optitrack

data were affected by the presence of the body suit,

its fitting to the participant, and by the location and

observability of the reflective markers. On the other

hand, the Kinect data were mainly affected by the

ambient lightning and by the lower resolution of the

sensors.

To check the reliability of the body model data,

the mean and STD of both the arm and forearm, as

identified by the two optical systems along the

acquired frames are calculated. It is shown that,

when the Kinect is located in front of the subject, the

deviations relative to the arm and the forearm

lengths computed by the Kinect were lower, as to

suggest that the device is more precise in such

configuration rather than in others, where it is

inclined respect to the subject. However, the results

obtained when the Kinect was in these latter

configurations do not dramatically exclude their

usability for a qualitative evaluation of the upper

limb motion. It was also observed that, if the upper

body of the subject is completely in the field of view

of the Kinect and no occlusion of the joints occurs

during motion, the body tracking allows to

approximate the trajectory of the wrist with lower

noise. However, the estimate of the joints position is

not accurate, as suggested by the STD of segments

length computed using the Kinect data. Indeed,

comparing the Kinect data with the Optitrack data,

as reported in Table 2, it is possible to note that, for

every acquisition step, the Optitrack identifies the

length of the upper arm with a very low STD,

whereas the Kinect gives different mean lengths for

each acquisition step. For the Optitrack data, the

mean length of the forearm results to be variable for

each acquisition step because of the motion tracking

suit, particularly at the wrist.

5 CONCLUSIONS

This preliminary study suggested that the Kinect

may be potentially adopted for applications

involving upper limb rehabilitation. The advantages

of such a system are the low cost, no requirement for

calibration, the easy to use and to set up, and the

absence of any body marker or suit that could

inevitably involve motion artefact. However, the

main limitation is the lower resolution, compared to

the more expensive marker-based systems.

Nevertheless, such resolution, of about few tens of

millimetres for the experimental test performed here

on the upper limb, make the Kinect an interesting

tool for applications in the rehabilitation field.

Future works are aimed at carrying out a more

detailed and extensive experimental analysis

involving several healthy subjects with different

characteristics, trying to come up with a more

reliable statistical analysis and some concluding

evidence on the usability of the device for the

specific body application.

REFERENCES

Asín Prieto G., Cano-de-la-Cuerda R., López-Larraz E.,

Metrot J., Molinari M., van Dokkum, L. E. H., 2014.

Emerging Perspectives in Stroke Rehabilitation. In

Kinect V2 for Upper Limb Rehabilitation Applications - A Preliminary Analysis on Performance Evaluation

137

Emerging Therapies in Neurorehabilitation

Biosystems & Biorobotics, vol. 4, pp 3-21.

SPRINGER.

Beucher, S., 1992. The watershed transformation applied

to image segmentation. In Scanning Microscopy-

Supplement-, pp 299-299.

Bonnechère, B., Jansen, B., Salvia, P., Bouzahouene, H.,

Omelina, L., Moiseev, F., Sholukha, V., Cornelis, J.,

Van Sint Jan, S., 2014. Validity and reliability of the

Kinect within functional assessment activities:

Comparison with standard stereophotogrammetry. In

Gait & Posture, vol. 39, no.1, pp 593 – 598.

ELSEVIER.

Burke, J. W., McNeill, M. D. J., Charles, D. K., Morrow,

P. J., Crosbie, J. H., McDonough, S. M., 2009.

Optimising engagement for stroke rehabilitation using

serious games. In The Visual Computer, vol.12, no. 25,

pp 1085-1099. SPRINGER.

Carse, B., Meadows, B., Bowers, R., Rowe, P., 2013.

Affordable clinical gait analysis: An assessment of

marker tracking accuracy of a new low-cost optical 3D

motion analysis system. In Physiotherapy, vol. 99, no.

4, pp347-351. ELSEVIER.

Chang,C.-Y., Lange, B., Zhang, M., Koenig, S., Requejo,

P., Somboon, N., Sawchunk, A.A., Rizzo, A.A., 2012.

Toword pervasive physical rehabilitation using

Microsoft Kinect. In 2012 6

International

Conference on Pervasive Computing Technologies for

Healthcare and Workshops, PervasiveHealth 2012, pp

159-162.

Clark, R.A., Pua, Y., Fortin, K., Ritchie C., Webster, K.E.,

Denehy, L., Bryant, A.L., 2012. Validity of the

Microsoft Kinect for assessment of postural control. In

Gait & Posture, vol. 36, no.3, pp 372-577.

ELSEVIER.

Diego-Mas, J. A., Alcaide-Marzal, J., 2014. Using

Kinect

sensor in observational methods for

assessing postures at work. Applied Ergonomics, vol.

45, no.4, pp 976 – 985. ELSEVIER.

Frisoli, A., Procopio, C., Chisari, C., Creatini, I.,

Bonofiglio, L., Bergamasco, M., Rossi, B.,

Carboncini, M. C., 2012. Positive effects of robotic

exoskeleton training of upper limb reaching

movements after stroke. In Journal of

Neuroengineering and Rehabilitation, vol. 9, no. 36.

BioMed Central.

Kinect for Windows features. Available from:

https://www.microsoft.com/enus/kinectforwindows/m

eetkinect/features.aspx [3August 2015]

Lachat, E., Macher, H., Mittet M. A., Landes, T.,

Grussenmeyer, P., 2015. First experiences with Kinect

v2 sensor for close range 3D modelling. In The

International Archives of the Photogrammetry,

Remote Sensing and Spatial Information Sciences,

Volum XL-5/W4.

Lam, P., Hebert, D., Boger, J., Lacheray, H., Gardner, D.,

Apkarian, J., Mihailidis, A., 2008. A haptic-robotic

platform for upper-limb reaching stroke therapy:

preliminary design and evaluation results. In Journal

of Neuroengineering and Rehabilitation, vol. 5, no. 15.

BioMed Central.

Lange, B., Chang, C., Suma, E., Newman, B., Rizzo, A.,

Bolas, M., 2011. Development and evaluation of low

cost game-based balance rehabilitation tool using the

Microsoft Kinect sensor. In Engineering in Medicine

and Biology Society, EMBC, 2011 Annual

International Conference of the IEEE, pp 1831-1834.

IEEE.

Mihelj, M., 2006. Human Arm Kinematics for Robot

Based Rehabilitation. In Robotica, vol. 24, no. 3, pp

377-383. CAMBRIDJE UNIVERSITY PRESS.

Moeslund, T. B., Hilton, A., & Krüger, V., 2006. A survey

of advances in vision-based human motion capture and

analysis. In Computer vision and image understanding,

vol. 104, no. 2, pp 90-126. ELSEVIER.

Natural Point, 2015. Inc.: Optitrack-optical motion

tracking solutions. Available from: https://

www.naturalpoint.com/.

Pimentel do Rosàrio, J.L., 2014. Biomechanical

assessment of human posture: A literature review. In

Journal of Bodywork and Movement Therapies,

vol.18, no. 3, pp 368-373. ELSEVIER.

Van Diest, M., Stegenga, J., Heinrich, J.W., Postema K.,

Verkerke, G. J., Lamoth, C. J. C., 2014. Suitability of

Kinect for measuring whole body movement patterns

during exergaming. Journal of Biomechanics, vol. 47,

no.12, pp 2925 – 2932. ELSEVIER.

Volpe, B. T., Lynch, D., Rykman-Berland, A., Ferraro,

M., Galgano, M., Hogan, N., Krebs, H.I., 2008.

Intensive Sensorimotor Arm Training Mediated by

Therapist or Robot Improves Hemiparesis in Patients

With Chronic Stroke. In Neurorehabilitation & Neural

Repair, vol. 22, no. 3, pp 305-310. SAGE.

Xu, X., McGorry, R. W., 2015. The validity of the first

and second generation Microsoft Kinect

for

identifying joint center locations during static

postures. In Applied Ergonomics, vol. 49, pp 47 – 54.

ELSEVIER.

Zhou, H., Hu, H., 2008. Human motion tracking for

rehabilitation-A survey. In Biomedical Signal

Processing and Control, vol.3, no. 1, pp 1-18.

ELSEVIER.

BIODEVICES 2016 - 9th International Conference on Biomedical Electronics and Devices

138