An Instrumented Glove for Swimming Performance Monitoring

M. Mangin

, A. Valade

, A. Costes

, A. Bouillod

2,4

, P. Acco

and G. Soto-Romero

1,2

ISIFC - Génie Biomédical, 23 Rue Alain Savary, Besançon, France

LAAS-CNRS, N2IS, Toulouse, France

Université de Toulouse, UPS, PRISSMH, Toulouse, France

EA4660, C3S - Université de Franche-Comté, Besançon, France

Keywords: Glove, Performance Monitoring, Embedeed Electronics, Swimming, Wearable Device.

Abstract: This paper presents a project of wearable motion capture system for motion analysis in swimming. Two

versions of this system have already been designed, one with a wired structure, based on a microcontroller

and an inertial measurement unit (IMU), and the other with a distributed architecture, based on a wireless

communication and another IMU. This system has been initially designed to target tri-athletes population,

but this study only presents the considerations concerning the swimming application.

1 INTRODUCTION

Movement efficiency is a challenge in the training of

swimmers in order to increase their performances

(Callaway, 2015; Psycharakis and Sanders, 2010;

Ohgi et al., 1998). A coach can easily measure

stroke frequencies or split times but currently, it is

difficult to evaluate the swimming technique.

Indeed, underwater, 2D or 3D cameras (Samson et

al., 2012) have been traditionally used in order to

collect swimming kinematics, but they are

cumbersome and expensive, and they require an

heavy post-processing and a correct brightness.

This study presents a part of a project which

consists on developing a smart electronic measuring

system, supposed to be wearable and composed of

multi-sensors areas communicating with a central

station and a computer. The main application is the

quantification of swimming kinematics. In order to

improve its wearability, we have chosen to integrate

it into a glove which can be used by the swimmers.

The purpose of this study was to validate

different parts of the future system (accuracy of the

system, waterproofness, wireless communication)

and to collect preliminary hand kinematics from

both elite and recreational swimmers.

2 ARCHITECTURE DESIGN

2.1 Wired Approach

Our first approach was based on a previous study

(Hernandez et al., 2014) which presented an

instrumented glove used for the capture of hand

gestures for a surgical application. We propose

another application in sports by firstly developing a

wired system monitoring the hand positions of a

swimmer. We used an inertial measurement unit

(IMU MPU9150, InvenSense), which uses Inter-

Integrated Circuit (I2C) communication standard.

The MPU9150 includes a 3-axis gyroscope with a

full-scale range of ±250, ±500, ±1000, and ±2000

°/sec, a 3-axis accelerometer with a full-scale range

of ±2, ±4, ±8, and ±16 g, and a 3-axis magnetometer

with a full scale range of ±1200 µT. The gyroscope

and the accelerometer are in 16-bit resolution, and

the magnetometer is in 13-bit resolution.

Figure 1: MPU9150 (left) and TIVA launchpad (right).

Mangin, M., Valade, A., Costes, A., Bouillod, A., Acco, P. and Soto-Romero, G..

An Instrumented Glove for Swimming Performance Monitoring.

In Proceedings of the 3rd International Congress on Sport Sciences Research and Technology Support (icSPORTS 2015), pages 53-58

ISBN: 978-989-758-159-5

This IMU was connected via wires to a

microcontroller on a Tiva C Series EK-

TM4C123GXL launchpad (Figure 1) which sent the

data at 50Hz. The microcontroller was programmed

to handle the data processing, which means reading

the raw data from the IMU, computing the sensor

orientation and sending the results to the computer

via USB data link.

2.2 Distributed Approach

Our second approach was to propose a distributed

processing (called AREM Gateway, or Gateway)

(Figure 2), where a sensor is connected to a

microcontroller (with embedded computing

algorithms) and equipped with a battery and a

wireless communication module. We made our

study with only one sensor, but our aim was to use

additional sensors. Thanks to this wireless

architecture, the Gateway would give two feedback

possibilities:

- a feedback at the end of a series of laps, with a

data post-processing,

- a direct feedback, as the data can be processed in

real-time by an algorithm embedded into the

microcontroller.

We used another IMU, a ST iNemo-M1, which

includes a 6-axis IMU (consisting on a 3-axis

accelerometer and a 3-axis magnetometer), a 3-axis

gyroscope and an ARM STM32 microcontroller. The

wireless transmission of the processed data is done

using an ESP8266 WiFi module working in a station

mode and connecting to a standard WiFi Access-

Point. We also added a USB connector for the

microcontroller, in order to communicate with a

computer and to charge the battery, and a Serial-

ATA connector connected to a Serial Peripheral

Interface (SPI) bus to enable extension capabilities

(with a SD card for example). Finally, the Gateway

is 31 mm wide, 44 mm long, 13 mm high (with the

battery and connectors), and weights 15 g with a 300

mAh battery, which has been tested and supply

power for 2h30 in normal operation mode.

Figure 2: Second version of the system (Gateway).

3 VALIDATION

3.1 Laboratory Tests

During the development of our systems, our first aim

was to validate the hand swimming kinematics

thanks to the IMU. In view of this, we compared

pitch, roll and yaw angles provided both by the IMU

of the wired system and the Gateway and a Vicon

system (using Nexus 1.7.1 software), a marker-based

motion capture system acknowledged as a reference.

This motion capture system carries 12 MX3+

cameras with a frequency of 200 Hz, a millimeter

accuracy and a resolution of 659 × 494 pixels each.

We set up two trials in order to compare the IMU

with the Vicon system. Each trial was filmed and

recorded with both systems (wired and distributed).

For the first, the IMU was surrounded with three

reflective passive markers (Figure 3) and controlled

by software (sample code from Texas Instruments,

and TeraTerm for the TIVA and a Python script for

the Gateway as computer softwares). We put it on a

table at the center of a room equipped with the

Vicon system, and we collected data with the IMU

and the Vicon system simultaneously during a

rotation about a spatial axis. We made these

rotations successively around the three axes in order

to obtain the roll, pitch and yaw movements. For the

second trial, we put the IMU on the hand of a

swimmer (Figure 4) and we asked him to simulate a

crawl movement while collecting data with both the

IMU and the Vicon system.

Figure 3: Settings for the first experiment (left: wired,

right: distributed).

Figure 4: Settings for the second experiment (left: wired,

right: distributed).

icSPORTS 2015 - International Congress on Sport Sciences Research and Technology Support

To compare the data collected with the IMU and

the Vicon system, we first had to convert the

coordinates provided by the Vicon cameras into

angles. To do so, we used an algorithm with a

simple angular projection leading to a coherent

result for simple rotations around X, Y or Z axis.

Finally, we were able to compare the data collected

by both systems (Figure 5).

Figure 5: Comparison between IMU and Vicon for the

roll, pitch and yaw angles (defined with respect to the

sensor’s orientation).

Moreover, we are currently working on an

algorithm based on the method described by Arun,

Huang and Blostein (Arun, Huang and Blostein,

1987), to determine rotation matrixes and translation

vectors with Vicon coordinates as inputs to deduce

Euler’s angles. Knowing the rotation matrix, we will

be able to determine the sensor’s attitude. This will

allow us to compare two methods of conversion, and

to determine which one is the most accurate.

3.2 Preliminary Field Tests

After the lab tests, we wanted to carry out field

validation tests. Ensuring waterproofness appeared

more difficult with the wired version, because we

hadn’t any glove prototype available yet.

Consequently, we decided to do the trials in pool

only with the Gateway.

Firstly, we tested the WiFi communication

provided by the ESP8266 WiFi module (2,4 GHz) in

a swimming pool in order to establish if the

communication was possible in such an environment

(water surrounding, metallic structure), and if it was

possible under water. To do so, we attached the

Gateway to the hand of a swimmer and we asked her

to put gradually her hand in the water until the loss

of the communication (Figure 6). The orientation of

the IMU is presented in Figure 7.

Figure 6: First tests of the WiFi communication.

Figure 7: IMU’s orientation during the field preliminary

tests.

We noted that the depth limit was about one

centimeter (under this depth, the WiFi signal was

lost), but that the communication was very good on

the surface of the water despite the environment

quite unfavorable to the correct travel of

electromagnetic waves.

Secondly, we decided to test the range of the

WiFi signal next to and in the swimming pool. We

noticed that the distance between the Gateway and

the surface of the water didn’t change the range of

the signal, and we determined it at about sixty

meters.

Thirdly, we tested data recording with the

Gateway during a fifty meters swim by two

swimmers. Because of the impossibility to

communicate underwater, we chose to record the

data when the WiFi signal was lost and to send them

at the end of each lap (at the communication

restoration). We managed to retrieve the data

corresponding to five crawl movements of each

swimmer.

4 RESULTS AND DISCUSSION

As the present study is a work in progress, the

following part will present results based on

preliminary field tests, introduce test protocols and

Yaw

Roll

Pitch

An Instrumented Glove for Swimming Performance Monitoring

analyses we want to realize with athletes, before

talking about the design we have proposed for the

glove.

4.1 WiFi Communication Tests

We draw out two options from this experiment: 1)

we had to change the frequency of the wireless

signal (in order to limit the absorption of the signal

by the water) or 2) to reconsider the wireless

communication strategy. In fact, we would have to

adapt our system to record the data and transmit

them to a computer when the Gateway is out of the

water (at the end of a swim lap). This improvement

would enable us to propose a real-time feedback, or

at least a faster feedback.

4.2 Swimming Evaluation

In order to get preliminary data related to the hand

movement of a swimmer, we will work in

collaboration with recreational and elite athletes.

We supposed that elite swimmers would have a

hand trajectory which permit them to be more

effective, that’s why it could be interesting to

compare hand kinematics of both groups in order to

assess differences between elite and recreational

swimmers.

Athletes will be asked to do a self-determined

warm-up before being equipped with the Gateway

on the left hand. Then they will have to perform

crawl during fifty meters, without a diving start, in a

fifty meters swimming pool. In order to standardize

the measurements, the swimmers will be asked to

put their hand at the surface of the water, in the

direction of the pool, before beginning to swim.

Data will be recorded in an external memory (SD

card) and collected on a computer at the end of each

lap.

4.3 Parameters Extraction

From the variations of pitch, roll and yaw angles

provided by the Gateway, the challenge was to

interpret the collected data. First, it is necessary to

correlate the curves obtained with hand positions,

and then to determine what is the most effective

trajectory. Moreover, it would be interesting to

evaluate if the hand acceleration (Hagema et al.,

2013) could be associated with swimming

effectiveness.

Our approach is the following: we are comparing

hand angles between different swimmers in order to

distinguish both elite and recreational swimmers.

Preliminary data comparing a recreational and an

elite swimmers are presented in Figure 8.

With the current state of the study, we can make

the following preliminary interpretations. A decrease

of the roll angle can be interpreted as the beginning

of the crawl movement, when the swimmer draws

water. In this part of the movement, the yaw angle

represents the hand’s direction. Indeed, if it

decreases, the hand is pulling from the left to the

right, and if it increases, the hand is pulling from the

right to the left. Afterwards, an increase of the roll

and yaw angles can correspond to a hand’s lateral

displacement, which is an example of a movement to

avoid.

A perspective of analysis is a comparison of two

crawl movements: a “correct” movement and a

movement where the arms are stretched (Figure 9).

Figure 8: Comparison between a recreational (top) and an

elite (bottom) swimmers.

Because of dealing with raw data in these

figures, we can emphasize the need of a fast filtering

algorithm to correct some unattended points like

“gimbal lock” and also compute useful data for real

time retrieval. Some recent works (Janota et al.,

2015) explain how to perform the Euler angles

computing, and we expect to include that correction

on IMU chips.

icSPORTS 2015 - International Congress on Sport Sciences Research and Technology Support

Figure 9: Two different crawl movements (top: incorrect,

bottom: correct).

However, our system also measures the 3-axis

accelerations, angular velocities (and magnetic field

which will not be detailed in this paper). A first

comparison between elite and recreational swimmers

showed a difference in acceleration amplitude on

different styles (Figures 10 and 11).

Figure 10: Acceleration (top, in g) and angular velocities

(bottom, in deg.s

-1

) during an elite swimmer test: 2x200m

frontcrawl and 200m Medley (red: X, green: Y and blue: Z

axis).

A more fine analysis showed differences on

patterns and amplitudes/frequencies for each style.

That suggests the interest of establishing a “quality

factor” qualifying the swim, based on pattern

analysis, to evaluate swimmer stroke and give a real

time feedback.

Figure 11: Acceleration (top, in g) and angular velocities

(bottom, in deg.s

-1

) during a recreational swimmer test:

200m frontcrawl and 200m Medley (red: X, green: Y and

blue: Z axis).

We show on next figure (Figure 12) the front

crawl pattern, being the most analyzed (Dadashi,

Crettenand and Millet, 2012). But that pattern will

be studied for other swimming styles, like

backstroke. We expect to be able to perform the

analysis of fatigue impact on swimmer stroke, by

storing data from long distance tests.

Figure 12: Comparison between elite (top) and

recreational (bottom) front crawl stroke pattern.

4.4 Glove Design

In parallel of the development of the electronic part,

we proposed different kinds of shapes for the glove.

For the prototype, we decided to develop a glove

without fabric on the palm and the fingertips. The

lengths of fabric on the fingers have been chosen in

order to permit the addition of pressure sensors

(Figure 13).

An Instrumented Glove for Swimming Performance Monitoring

Figure 13: Drawings of the first design of the glove.

The glove has been proposed to two swimmers to

collect their perceptions and their opinions. They

appreciated the fact that the palm and the fingers tips

were free, and didn’t feel disturbed by the glove. But

they underlined the need to adjust correctly the size

in order to prevent the passage of water at the back

of the hand, and the slight difficulty of putting the

glove on.

5 CONCLUSIONS

The preliminary interpretation seems to show that

the Euler’s angles variation would be a first

interesting parameter to quantify the swimming

technique. Indeed, the correlation between the

curves and the movements would enable to provide

to the swimmer certain necessary adjustments

without a video recording or the observation of his

coach. But the understanding of the curves obtained

implies a correction of the data collected, because

the curves seem to contain some aberrant points.

A more common analysis, according to literature,

shows a difference in patterns between elite and

recreational swimmers. More than recognizing the

swimming style, the next step will be to extract a

quality stroke index by style and perform fatigue

analysis on swimmers.

This study represents a step forward in the

development of a wearable motion capture system to

monitor swimming performances. It also underlines

the fact that the wireless communication must be re-

engineered in order to transmit data underwater.

In a further study, we would like to add several

pressure sensors in order to provide information

relative to the force exerted by the athlete on the

water. This would be another parameter to assess

swimming kinetics, in addition to the kinematics

provided by the sensors presented in this study.

ACKNOWLEDGEMENTS

This research project is partially supported by TE

Connectivity and Compressport International. The

authors would like to thank Mrs. Marie Percebois,

Mr. Manuel Roux, Mr. Louis Cupillard, and others

coaches and athletes from Besançon Triathlon and

Toulouse Université Club (TUC Triathlon), for their

assistance in swimming tests.

The authors would also like to gratefully

acknowledge Mrs. Kris Martinez, Mr. Xavi Carabi

and Mr. Sylvain Laur, from Compressport Int, for

their help in glove design and first prototypes of

wearable devices.

REFERENCES

Callaway, A. J. Measuring Kinematic Variables in Front

Crawl Swimming Using Accelerometers: A Validation

Study. Sensors 2015, 15, 11363-11386.

Psycharakis, S. G.; Sanders, R. H. Body roll in swimming:

A review. J. Sports Sci. 2010, 28, 229–236.

Ohgi, Y.; Ichikawa, H.; Miyaji, C. Characteristics of the

forearm acceleration in swimming. Proceedings of the

Symposium on Biomechanics and Medicine in

Swimming VIII, Jyväskylä, Finland, 28 June–2 July

1998; Kesikinen, K.L., Komi, P.V., Hollander, A.P.,

Eds.; pp. 77–82.

Samson, M.; Bernard, A.; David, L. Mesures cinématiques

et dynamiques en fin de phase d’allongement du

crawl. 13

ème

Congrès Francophone de Techniques

Laser 2012.

Hernandez, A. M.; Lemos, J. D.; Soto-Romero, G.;

Valade, A. Instrumented glove for skills assessment in

neurosurgical simulation. Biomedical Circuits and

Systems Conference (BioCAS), 2014.

Arun, K. S.; Huang, T. S.; Blostein, S. D. Least-squares

fitting of two 3-D point sets. IEEE Transactions on

Pattern Analysis and Machine Intelligence, 9, 5, 1987,

pp. 698-700.

Hagema, R. M; Haelsig, T.; O’Keefe, S. G; et al. Second

generation swimming feedback device using a

wearable data processing system based on underwater

visible light communication. Procedia Engineering

60, 2013, pp. 34-39.

Aleš Janota; Vojtech Šimák; Dušan Nemec and Jozef

Hrbček. Improving the Precision and Speed of Euler

Angles Computation from Low-Cost Rotation Sensor

Data. Sensors 2015, 15, 7016-7039.

Farzin Dadashi; Florent Crettenand; Grégoire P. Millet

and Kamiar Aminian. Front-Crawl. Instantaneous

Velocity Estimation Using a Wearable Inertial

Measurement Unit. Sensors 2012, 12, 12927-12939.

icSPORTS 2015 - International Congress on Sport Sciences Research and Technology Support