A Robot-based Application for

Physical Exercise Training

Michalis Foukarakis, Ilia Adami, Danae Ioannidi, Asterios Leonidis, Damien Michel,

Ammar Qammaz, Konstantinos Papoutsakis, Margherita Antona and Antonis Argyros

Foundation for Research & Technology – Hellas (FORTH), Institute of Computer Science,

Vassilika Vouton, 70013 Heraklion, Greece

Keywords: Physical Exercise, Elderly People, Assistive Robot, Human Motion Capture, Gesture Recognition.

Abstract: According to studies, performing physical exercise is beneficial for reducing the risk of falling in the elderly

and prolonging their stay at home. In addition, regular exercising helps cognitive function and increases pos-

itive behaviour for seniors with cognitive impairment and dementia. In this paper, a fitness application inte-

grated into a service robot is presented. Its aim is to motivate the users to perform physical training by provid-

ing relevant exercises and useful feedback on their progress. The application utilizes the robot vision system

to track and recognize user movements and activities and supports multimodal interaction with the user. The

paper describes the design challenges, the system architecture, the user interface and the human motion cap-

turing module. Additionally, it discusses some results from user testing in laboratory and home-based trials.

1 INTRODUCTION

Physical well-being affects a great deal the overall

quality of life of any person, but even more so that of

a senior, as it may be one of the determining factors

for allowing that person to continue living an inde-

pendent life at home. The multiple positive effects of

regular physical exercise on maintaining and improv-

ing the overall physical and mental well-being of the

elderly are well known and documented. Physical fit-

ness is associated with better balance, motor skills,

and improved reaction times for the elderly (Buchner

et al, 1997). As data acquired from multiple studies

have demonstrated (Laurin et al., 2001; Heyn et al.,

2004), it has also been associated with a lower risk of

developing cognitive disorders like dementia. Im-

proving upper limb function and increasing trunk en-

durance and trunk strength has been proven to im-

prove balance (Suri et al., 2009) and stabilize posture

(Kaminski et al., 1995), thus reducing the risk of falls.

However, in order for any exercise routine to be ben-

eficial to the elderly, it is important that it is designed

specifically for their needs, while taking into account

all the potential physical limitations that may be pre-

www.hobbit-project.eu

sent due to an ailment or due to age related dysfunc-

tions (e.g., limited flexibility of the joints, shorter

ranges of motion, etc.). In addition, supervision by a

trainer is essential for ensuring that the exercises are

executed correctly, thus limiting the risk of injury,

and for providing appropriate encouragement or ad-

ditional instructions when necessary. Many nursing

and care facilities for the elderly offer exercise classes

instructed by trainers. However, access to such struc-

tured classes is not always easy or possible for elderly

people for many reasons, such as lack of transporta-

tion, inability to leave the house easily due to health

conditions, geographic isolation, schedule conflicts,

etc.

A potential solution to this problem has been ex-

plored in the context of the EU funded project

HOBBIT

. For this purpose, a fitness application has

been designed and developed, which is integrated in

the HOBBIT robotic platform for the elderly to use

on demand. More specifically, this application is a

guided fitness routine especially designed for the

needs of the elderly that can be practiced in a safe and

pleasant way. The application runs as a component of

the HOBBIT robot system, a low cost mutual care as-

sistive robot targeted to reduce the risk of falling at

home through a variety of functions.

Foukarakis, M., Adami, I., Ioannidi, D., Leonidis, A., Michel, D., Qammaz, A., Papoutsakis, K., Antona, M. and Argyros, A.

A Robot-based Application for Physical Exercise Training.

In Proceedings of the International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2016), pages 45-52

ISBN: 978-989-758-180-9

2 RELATED WORK

In recent years, researchers have explored ways to ex-

ploit the technological advances of gesture recogni-

tion and motion tracking systems in the field of reha-

bilitative and physical therapy, with very positive re-

sults. Virtual reality gaming using commercially

available systems, such as Microsoft’s Kinect and

Nintendo’s Wii platform, has been shown to improve

measures of upper limb function in stroke survivors

and balance in elderly people in community and hos-

pital settings (Davies et al., 2014). A quick literature

review on the subject of Kinect and physiotherapy

will produce many results of such research projects

which aim to develop applications that can provide

real time virtual assistance to patients performing spe-

cific exercises, while also helping therapists to track

and monitor the progress of their patients even from

a remote location (Saposnik & Levin, 2011; Chang et

al., 2011; Wei-Min et al., 2013). The subject of ro-

botic assistants that facilitate physical training is ad-

dressed in (Fasola & Mataric, 2012; Werner et al.,

2013).

Systems in these studies as well as commercial ap-

plications use a similar setting; the user stands in front

of a large screen that displays two avatars, one that

represents the trainer and the other the user. The mo-

tion-tracking camera reads the movements of the user

and transfers them on the respective avatar. In most

of these systems, the movements of the patient are

displayed on the screen, against or in parallel to the

movements of the trainer avatar. In that way, the user

has immediate visual feedback on how accurate his or

her movements are compared to the ones demon-

strated by the trainer avatar. In other systems, the user

has to perform specific movements in order to score

points in the context of a fun video game. Using video

gaming platforms in physical therapy has been shown

to be an effective method that increases the engage-

ment and motivation of patients to follow and com-

plete their physical program (Davies et al., 2014; Un-

nikrishnan et al., 2013).

3 DESIGN CHALLENGES

The main objective of the fitness application is to pro-

mote and facilitate physical fitness for the users of the

robot in their own home environment. To achieve

this, the application provides a guided exercise pro-

gram especially designed for the elderly, which can

be initiated from the robot’s main interface at any

time. The application is also able to track the user’s

performance during the exercise and provide con-

structive feedback when needed. The application is

not to be intended as a therapeutic or rehabilitation

tool, but as a tool to help the users maintain or im-

prove their motor skills and balance. The application

offers exercises that are presented sequentially

through a video trainer guiding the user throughout

the fitness program. It uses a human tracking system

in order to detect and display user’s movements,

through the avatar, while executing an exercise. The

user’s avatar is depicted right next to the virtual

trainer for direct visual comparison. Feedback to the

user (i.e., instructions, corrections, and general en-

couragement), is provided through HOBBIT in a mul-

timodal manner, auditory and textual.

3.1 User Requirements

The user requirements for the fitness application were

gathered and elaborated through brainstorming and

observation sessions at a local day-care and activity

center for the elderly, with the involvement of geriat-

ric occupational therapists and HCI experts. The cen-

ter offers a 30 minutes fitness program twice a week,

where a group of 25 members have the opportunity to

exercise the upper body while seated. The elaboration

of the user requirements was based on two videotaped

sessions presenting the potentials of the elderly in ex-

ecuting upper body exercises, on information gath-

ered through an anonymous short questioner, as well

as on valuable brain-storming sessions with the elder

members and the center personnel.

The following requirements were derived:

• Overall Experience. The application must be

pleasant and engaging.

• Usable Interaction Paradigm. The user interface

should be designed in a way that the interaction

paradigm is close to that of a real fitness class set-

ting,

• Continuous Flow of Interaction. The user should

not wait for long periods of time between exercises

or application dialogues to avoid frustration or

even fatigue.

• Designated Exercise Area and Position. All the

exercises must be performed from a seated posi-

tion. Exercising from a sitting position improves

safety, since the base of support is greater than that

of standing, while also increases the ability to per-

form motor activities with arms and hands (Greene

& Roberts, 2005). The user’s chair must be sturdy

with no bulky legs and arms to allow uncon-

strained body movements and avoid tracking mis-

reading and miscalculations. An area in front of the

ICT4AWE 2016 - 2nd International Conference on Information and Communication Technologies for Ageing Well and e-Health

user with appropriate space for the robot to stand

is also needed.

• Elderly Specific Exercise Routines. The exercise

routines must be designed for the average 70+

years of age user that is not experiencing any major

or fitness forbidding health and physical problems.

The exercise routines should be designed in close

consultation with health professionals who have

expertise for the elderly (e.g. occupational and

physical therapists).

• Instructions. The instructions given by the trainer

avatar must be precise, short, and clear.

• GUI Visibility. The avatars must be large enough

to be clearly seen from required 2 meters distance

that the user has to keep from the camera sensor of

the robot.

3.2 Technical Constraints

The design of the fitness application had to overcome

various challenges that derived from limitations and

constraints of the specific robot-based set-up. These

include:

• Operation Distance. HOBBIT uses a Microsoft

Kinect visual sensor

to detect the user and track

his/her movements. This means that the sensor re-

quires the user to be within its field of view and at

a certain distance in order to work properly. As a

consequence, the robot needs to stand approxi-

mately 2 meters away from the user at all times

during the execution of the fitness application. Fig-

ure 1 shows the recommended user/robot setup.

Figure 1: The fitness application setup. HOBBIT is stand-

ing at least two meters in front of the user.

• Screen Size. HOBBIT has a tablet on the front side

where the application is shown. The screen’s diag-

onal size is 12.1 inches, which is considered rather

small for long distance viewing and poses signifi-

cant constraints on the user interface design.

https://dev.windows.com/en-us/kinect

• Limited Interaction. The required distance be-

tween the robot and the user renders the touch

screen interaction method and textual feedback in-

efficient, making other modes of interaction (i.e.,

gestures, voice commands, verbal output) essential

for communication.

• Sensing Constraints. Despite the fact that state-

of-the-art, patent pending technology is used to

support human motion tracking and activity recog-

nition (Michel & Argyros, 2014), certain body

poses cannot be recognized (e.g., side views that

lead to severe self-occlusions). Therefore, the ex-

ercise routine should exclude such poses to avoid

human pose estimation errors.

• System Complexity. HOBBIT is a complex sys-

tem that provides multiple functionalities which

need orchestration. The fitness application is only

part of this system and needs to cooperate fully

with the rest of the components. In most cases

(emergencies, low battery power, etc.) the applica-

tion needs be able to terminate gracefully and

transparently to the user.

4 SYSTEM ARCHITECTURE

AND USER INTERFACE

Figure 2 shows the architecture of the fitness applica-

tion, its internal components and its communication

with the overall HOBBIT system. The application is

divided into three modules: (i) The graphical user in-

terface (GUI), which takes care of the visual part of

the application and accepts user input when required;

(ii) the sensor module, which is responsible for rec-

ognizing user movements and tracks the user skeleton

so that it can be shown in the user interface; (iii) the

controller module, which acts as an intermediate

communication layer between the other two modules.

The controller module acts as an intermediary

among the other components. It is responsible for

translating and passing any messages that come from

the sensor module to the GUI module and vice versa.

This data includes information about which exercise

is currently tracked, if a repetition was successful or

not, if an error has occurred, and detected user skele-

ton joint coordinates. It also receives messages from

HOBBIT’s multimodal User Interface (MMUI) to fa-

cilitate robust integration with the rest of the user in-

terface system.

The sensor module is the one responsible for de-

tecting specific (correct or erroneous) user move-

A Robot-based Application for Physical Exercise Training

ments while he or she is doing an exercise. To this

end, four specific physical exercises were designed

for the elderly, including repetitive movements for

lateral or/and bilateral abduction (or 90

elevation) of

shoulders with upper extremities extended and lateral

and bilateral elbow flexion with shoulder elevation in

. The exercises follow instructions and examples

from the official website of the National Health Ser-

vice in England

. This module is described in more

details in section 5.

The GUI module is responsible for presenting au-

diovisual information to the user based on the ap-

proach described below.

Each of the exercises is demonstrated to the user

through a video displayed on the touch screen (tablet)

of the platform while the platform is placed towards

the user at a distance of at least 2 meters. HOBBIT

prompts the user to execute a specific number of rep-

etitions for a specific exercise providing auditory

guidance. During exercise execution, the GUI is split

in the following 3 major areas: (i) Trainer area where

the screen shows a video presenting a human trainer

executing the exercise; (ii) User area where the user

avatar (a visualization of current body configuration

of the observed user) is shown; (iii) Feedback area

displaying messages in the bottom center.

The system is able to recognize the repetitive

movements of the arms corresponding to the known

ongoing exercise and as a result to count them as ex-

ecuted repetitions. Moreover, any errors on the exe-

cution of an exercise compared to its known model

are evaluated in order to provide relevant feedback to

the user on the quality of his/her performance and po-

tential correction to be applied in the next round of

repetitions. For example, one of the four exercises re-

fers to six repetitions of left arm movement from ver-

tical to horizontal position and back to vertical. Some

type of errors that can be assessed involve, usage of a

wrong arm (e.g. the right arm instead of the left one)

of the user or the arm should be extended closer to the

horizontal position or the arm should be lowered

closer toward the vertical position, etc. Auditory feed-

back toward the user can be provided at the time of

execution or after the completion of the exercise. If

the perceived repetitions were completed without any

errors, the user will receive a positive message; oth-

erwise relevant auditory feedback is provided letting

the user know what kind of movement error was as-

sessed during the repetitions.

http://www.nhs.uk

Figure 2: Fitness application communication architecture.

5 HUMAN MOTION CAPTURE

AND GESTURE RECOGNITION

The developed fitness application relies on a com-

puter vision method (Michel & Argyros, 2014) that

estimates the 3D position, orientation and full articu-

lation of the upper human body from markerless vis-

ual observations obtained by an RGBD camera. This

method takes into consideration the high dimension-

ality and the variability of the tracked person regard-

ing appearance, body dimensions, etc. It performs un-

obtrusive, markerless tracking that does not interfere

with the environment, the subject and its actions. The

method achieves real time performance on a conven-

tional computer and relies on an inexpensive sensory

apparatus (RGBD or depth camera). It exhibits ro-

bustness in a number of challenging conditions (illu-

mination changes, environment clutter, camera mo-

tion, etc). In contrast to other methods, it performs au-

tomatic human detection and automatic tracking ini-

tialization and offers high tolerance with respect to

variations in human body dimensions, clothing, etc.

The interaction of the user with the robot is sup-

ported by vision-based gesture recognition (Michel et

al., 2014). The considered gestural vocabulary con-

sists of five user specified hand gestures that convey

important messages in the context of human-robot di-

alogue. Despite their small number, the recognition of

these gestures exhibits considerable challenges. Aim-

ing at natural, easy-to-memorize means of interac-

tion, users have identified gestures consisting of both

static and dynamic hand configurations that involve

different scales of observation (from arms to fingers)

and exhibit intrinsic ambiguities. The gestures are

recognized regardless of the multifaceted variability

of the human subjects performing them. Recognition

ICT4AWE 2016 - 2nd International Conference on Information and Communication Technologies for Ageing Well and e-Health

is performed online, in continuous video streams con-

taining other irrelevant/non-modeled motions. All the

above are achieved by analyzing information ac-

quired by the possibly moving RGBD camera, in clut-

tered environments with considerable light variations.

The invariance to light variations is due to the fact that

gesture recognition is based solely on the depth infor-

mation provided by the RGBD cameras. Several in-

stances of RGBD cameras have been tested success-

fully including Kinect and Xtion (structured light

emission) as well as Kinect 2 (time of flight).

5.1 Methodology

A 3D skeletal-based body model is used to estimate

body pose and track body parts, providing the re-

quired representation to model any observed body

configuration for each frame. A common exercise

model for the fitness application has been defined that

incorporates the following data fields: (i) a collection

of sequential 3D body configurations that correspond

to manually selected key-point configurations of the

physical movements for an ideal execution of the ex-

ercise; (ii) a unique label; (iii) the expected number of

iterations of the complete exercise; (iv) maximum al-

lowed time intervals between consecutive key-point

body configurations of an exercise; (v) Maximum al-

lowed duration for a complete exercise to be exe-

cuted.

The type of supported exercises and the key-point

body configurations of each of them have been man-

ually selected with the help of physiotherapists and

physical trainers. The experts have indicated a valid

set of appropriate, meaningful and safe body config-

urations to be considered as key-points for an exercise

and the transition between them with respect to phys-

ical conditions of elderly. Moreover, reasonable max-

imum allowed time intervals between consecutive

key-point body configurations have also been set by

the experts.

An example of a simple fitness exercise, that is

extending the arms from a seated idle body pose, is

demonstrated in figure 3, based on its key-point body

configurations. A pose-similarity function has been

defined to evaluate the proximity between two body

configurations, given the 3D positions and angles of

the tracked body joints with respect to the detected

body center, extracted by the human detection and

tracking module. The similarity score is defined

based on a weighted sum between the sum of absolute

3D Euclidean distances and the sum of absolute dif-

ferences of angles between corresponding joints

given that the two body centers are aligned. It is max-

imized in case the 3D positions and angles of the

tracked joints coincide with the corresponding joints

of the exercise model. Moreover, the absolute differ-

ences between corresponding body joints is stored for

subsequent use. The defined human posture similarity

metric is invariant to different user body sizes and di-

mensions as joint positions are normalized with re-

spect to the body center and the relevant body dimen-

sions.

The final step is to assess the similarity between

two activities in terms of body configurations pro-

vided by the exercise model to be executed, given as

input by the GUI module of the application, and the

time series of body configurations computed by the

vision-based user observation module. To this end, an

activity similarity function has been defined. It calcu-

lates a confidence value reflecting which of the ob-

served body configurations were highly similar to

Figure 3: An example fitness exercise is demonstrated

based on its three key-point body configurations.

A Robot-based Application for Physical Exercise Training

which of the exercise key-point body configurations,

how similar they were based on the pose-similarity

score and if the temporal order of those matches fol-

lows the requirements set by the exercise model. The

time intervals between matched body configurations

are also calculated and taken into account.

The activity similarity score for an exercise and

the time intervals between matched key-point config-

urations constitute the output of the developed mod-

ule. Moreover, statistics on the time series of absolute

residuals between corresponding skeletal joints in or-

der to identify those joints that were not matched

properly to the positions of the model are calculated.

Therefore, it is assumed that those joints are mainly

responsible for the possible failure to perform appro-

priately the intended exercise.

The output information facilitates the UI control-

ler module of the application to select and provide ap-

propriate instructions and prompts to the user as guid-

ance or feedback in order to repeat an exercise and

learn it successfully.

6 ITERATIVE DEVELOPMENT

AND USER TESTING

The application went through various iterative devel-

opment phases. The first priority was to establish ro-

bust communication architecture as described in sec-

tion 4. After that, gradual improvements were made

to the user interface and the controller module to ac-

commodate user comments, usability concerns, sen-

sor data changes and bug fixes.

A preliminary version of the fitness application

has been tested both in the lab and in a user’s home

environment (see figure 4).

The first lab tests were performed by interaction

and user interface experts and an older adult user. The

users were instructed to activate the fitness option

from the robot’s menu and follow the instructions.

The program included three exercises where the user

was prompted to raise first the left arm to the level of

the shoulder, then the right arm and finally both arms.

Figure 4: User testing the first application prototype.

From the feedback received, a number of changes

were deemed necessary. First, a new trainer video

was recorded, shortening the introduction and remov-

ing the lengthy trainer demonstration of each exercise

to reduce distractions and ensure that the user will

start training immediately and will not be confused

about when to start. Second, exercise feedback was

redesigned to be presented immediately after the user

has done something that is detected as either wrong

or deviates from the correct exercise form. This way,

the user will have instant feedback on errors that

he/she is doing during the exercise. Third, the user

avatar was made more visible and as a mirror to the

user’s movements. Trainer instructions as to which

arm is going to be used during each exercise, were

added in order to eliminate user confusion. Fourth,

two more arm exercises and two more exercise pro-

grams were added. The three exercises of the first

prototype application were included in the ‘easy’ pro-

gram, while two more programs, ‘medium’ and ‘hard’

were added, using combinations of the available ex-

ercises. The three new programs had different num-

bers of repetitions to allow more variations and avoid

repetitive sessions and additional resting intervals to

prevent user fatigue. Finally, the repetition counter in

the user interface was found to be too small, thus it

was redesigned to be larger and more visible.

This new version of the fitness application was

tested again by lab personnel and by two elderly users

in their home environment. This improved version

was better accepted by the elderly users. They gave

positive comments about the overall application per-

formance and layout. One of the complaints was that

the trainer feedback was unnecessarily repetitive and

frequent. To improve the user experience, the fre-

quency and content of this feedback were adjusted ac-

cordingly. Some examples of the fitness application in-

tegrated into HOBBIT’s MMUI are shown in figure 5.

Finally, the fitness application has been tested in

the context of the overall home-based trials of the

HOBBIT robot, involving a total of 18 users of age

75+, with various degrees of vision, hearing and mo-

bility impairments, in 3 European countries (Austria,

Sweden and Greece). Each user had a prototype

HOBBIT robot at home for a period of three weeks

and could freely use all functionality of the robot, in-

cluding the fitness application. Results were obtained

through a mix of quantitative and qualitative meth-

ods, cultural probes and logging. Regarding the fit-

ness function, the results were mainly collected

through interviews with the users. 17 users said that

they had tried the fitness function. The study revealed

diversity with regard to how the fitness feature was

perceived, with the Greek users being more satisfied

than the users in Austria and Sweden. However, the

ICT4AWE 2016 - 2nd International Conference on Information and Communication Technologies for Ageing Well and e-Health

results confirm that fitness was perceived by the users

in all countries as an important feature, pointing to the

need for wider selection of fitness exercises for a ro-

bot for the elderly. One participant who did not use

the fitness function had severe visual impairment and

was unable to read text on the icons or see what the

fitness instructor did when the robot was placed 2m

away from her. This points to the fact that the elderly

have very diverse needs and the interface and func-

tionality of an assistive robot needs to be adapted to

the individual user and his or her needs. Finally, ac-

cording to their comments, the users enjoyed the fit-

ness function, and some mentioned that their level of

exercise increased when they had HOBBIT due to the

fitness program.

Figure 5: Final version of the application.

7 CONCLUSION AND FUTURE

WORK

This paper has presented a physical exercise applica-

tion meant to be used by the elderly in the context of

a mutual care robot developed by the HOBBIT pro-

ject. The paper has presented the design process, the

user requirements, the technical constraints, the hu-

man motion capture and gesture recognition methods

and the development and testing process of the fitness

application. The need to develop this application was

originated from the user requirements of the HOBBIT

project. One of the goals of the project was to prevent

falls and apart from some of the robot’s functionali-

ties implemented for this purpose, a fitness applica-

tion was suggested as a means for fall prevention.

Future improvements include user interface adap-

tation features to increase accessibility and usability,

but also motivation and confidence. The following

types of adaptation are being considered: (i) suggest

the appropriate level of exercise based on past activ-

ity; (ii) adaptive motivational messages based on

tracked performance; (iii) adaptive feedback based on

success rate when performing an exercise; (iv) per-

sonalized information at the end of each exercise sec-

tion; (v) Adaptable pause behavior including pause

and automatically resume on external intervention

(e.g., phone call) and self-termination on long inac-

tivity; (vi) Identify inactivity and motivate participa-

tion (during exercises).

In addition to being integrated in the HOBBIT

platform, the fitness application can potentially also

be used independently as a standalone application in

older adult care facilities or in homes. The available

exercises focus on upper-body tracking of a seated

user. In an improved future version, a wider range of

exercises can be programmed, including leg exercises

for both seated and standing users. The user interface

can also be redesigned and improved since the robot’s

limitations (e.g. small screen size, inexpensive sen-

sors) would not apply in this case.

ACKNOWLEDGEMENTS

Part of this work has been conducted in the context of

the Project ICT-HOBBIT “HOBBIT The Mutual

Care Robot”, funded by the European Commission

under the 7th Framework Programme (Grant Agree-

ment 288146). The authors would like to thank the

TALOS Community Care & Active Ageing Center of

Heraklion Municipality for their significant contribu-

tion to the selection of the exercise program.

REFERENCES

Buchner, D.M., Cress, M.E., de Lateur, B.J., Esselman,

P.C., Margherita, A.J., Price, R., Wagner, E.H.: The Ef-

fect of Strength and Endurance Training on Gait, Bal-

A Robot-based Application for Physical Exercise Training

ance, Fall Risk, and Health Services Use in Commu-

nity-living Older Adults. Journals of Gerontology Se-

ries A-Biological Sciences & Medical Sciences, 52(4),

M218-24 (1997).

Chang, Y.J., Chen, S.F., Huang, J.D.: A Kinect-based Sys-

tem for Physical Rehabilitation: A Pilot Study for

Young Adults with Motor Disabilities. Res. Dev. Disa-

bil 32, 2566–2570 (2011).

Davies, T. C., Vinumon, T., Taylor, L., Parsons, J.: Let's

Kinect to Increase Balance and Coordination of Older

People: Pilot Testing of a Balloon Catching Game. In-

ternational Journal of Virtual Worlds and Human-

Computer Interaction. Vol. 2, 37-46 (2014).

Fasola, J., Mataric, M. J.: Using Socially Assistive Human–

robot Interaction to Motivate Physical Exercise for

Older Adults. Proceedings of the IEEE 100(8), 2512-

2526 (2012).

Greene, D.P., Roberts, S.L., 2005. Kinesiology: movement

in the context of activity, Elsevier Mosby. St. Louis, 2

edition.

Heyn, P., Abreu, B. C., Ottenbacher, K. J.: The Effects of

Exercise Training on Elderly Persons with Cognitive

Impairment and Dementia: A Meta-analysis. Archives

of physical medicine and rehabilitation 85(10), 1694-

1704 (2004).

Kaminski, T.R., Bock, C., Gentile, A.M.: The coordination

between trunk and arm motion during pointing move-

ments. Experimental Brain Research. 106(3), 457-466

(1995),

Laurin D., Verreault R., Lindsay J., MacPherson K., Rock-

wood K.: Physical activity and risk of cognitive impair-

ment and dementia in elderly persons. Arch Neurol 58,

498-504 (2001).

Michel, D., Argyros, A.: Apparatuses, methods and systems

for recovering a 3-dimensional skeletal model of the hu-

man body, U.S. provisional patent application No.

62/053,667, 22/09/2014.

Michel, D., Papoutsakis, K., Argyros, A.: Gesture Recogni-

tion Supporting the Interaction of Humans with So-

cially Assistive Robots. Advances in Visual Computing

- 10th International Symposium, ISVC 2014, 793-804,

Springer (2014).

Saposnik, G., Levin M.: Virtual Reality in Stroke Rehabil-

itation: A Meta-analysis and Implications for Clini-

cians. Stroke 42, 1380–1386 (2011).

Suri P., Kiely D.K., Leveille S.G., Frontera W.R., Bean

J.F.: Trunk muscle attributes are associated with bal-

ance and mobility in older adults: a pilot study. PM R.

1, 916–924 (2009).

Unnikrishnan, R., Moawad, K., Bhavani, R. R.: A physio-

therapy toolkit using video games and motion tracking

technologies. In: Global Humanitarian Technology

Conference: South Asia Satellite (GHTC-SAS), pp. 90-

95, IEEE (2013).

Wei-Min, H., Chih-Chen, C., Shih-Chuan W., Yu-Luen C.,

Yuh-Shyan H., Jin-Shin L.: Combination of the Kinect

with Virtual Reality in Balance Training for the Elderly.

Engineering 5, 171-175 (2013).

Werner, F., Krainer, D., Oberzaucher, J., Werner, K.: Eval-

uation of the Acceptance of a Social Assistive Robot for

Physical Training Support Together with Older Users

and Domain Experts. Assistive Technology: From Re-

search to Practice, band 33, 137-142, IOS Press (2013).

ICT4AWE 2016 - 2nd International Conference on Information and Communication Technologies for Ageing Well and e-Health