Freestyle Swimming Analysis of Symmetry and Velocities using a

MEMS based IMU: Introducing a Symmetry Score

Andy Stamm

1,2 a

and Igor Shlyonsky

Faculty of Technology and Bionics, Rhine-Waal University of Applied Sciences, 47533 Cleve, Germany

Griffith School of Engineering, Griffith University, 4111 Nathan, QLD, Australia

MySwimEdge, CA, U.S.A.

Keywords: Swimming, Freestyle, Symmetry, Velocity, IMU, Freestyle Symmetry Score.

Abstract: MEMS sensors (IMU’s) are widely available nowadays and tend to be used more often in sports monitoring.

Especially in swimming these sensors have seen rapid development in the past years. These sensors have very

good measurement capabilities today, but the automatic analysis of the gathered data has not yet been

implemented. Our objective is to develop and validate an automatic analysis which can provide the

swimmers/coaches with nearly immediate feedback on a smartphone/tablet. Ten swimmers ranging from

novice to elite have been participating in this study performing freestyle in either 25m or 50m pools. All trials

were recorded with a 3-axis accelerometer. The symmetry parameters have been extracted from the recorded

data after these were high-pass filtered to remove the gravity from the signal and a zero crossing detection

algorithm was applied. The results showed a very strong relation to results obtained by other researchers.

1 INTRODUCTION

Since many years, the performance of athletes in

swimming was evaluated by coaches sometimes

under the help of bulky and complex equipment such

as (multi) video camera systems and/or tethered

velocity meters (Craig & Pendergast, 1979; Craig,

Termin, & Pendergast, 2006; Stamm, Thiel, Burkett,

& James, 2009). Operating such equipment usually

needs a special trained person and additionally one

expert for the data analysis, thus not allowing using

this equipment on a regular basis. Furthermore a

tethered device allows only investigating the

movement into one direction; namely only one

swimming lap at a time. This leads to athletes/coaches

not using this equipment very often.

Nowadays athletes have become too competitive and

sometimes a tenth of a second can decide upon

gaining the next better place (Dadashi, Millet, &

Aminian, 2013; Magalhaes, Vannozzi, Gatta, &

Fantozzi, 2015) thus pushing the needs to monitor

basically each training session or competition.

https://orcid.org/0000-0002-9331-7431

Inertial Measurement Units (IMU’s) have become

smaller in size and lighter in weight in recent years,

allowing using such devices without any disturbance

and performance problems. These devices are

nowadays waterproof, easy to use (can be placed by

the athlete) and are able to record multiple training

sessions (Callaway, 2015; Guignard, Rouard,

Chollet, & Seifert, 2017; Stamm & Thiel, 2015).

IMU’s can nowadays be used to find key factors such

as stroke rate, split times, mean velocity, and arm

symmetry. The last one has only been presented by

(Stamm & Thiel, 2015) and is still novel in swimming

research purely based on IMU’s.

This research used a sacrum mounted MetaMotionC

IMU (mbientlab, 2020) packaged in a waterproof

casing to find the 3-axis acceleration dynamics of the

swimmer. These data were used for automatic

processing to find symmetry variables for further

investigations and analysis. The objective of this

research was to develop and validate an automatic

analysis with the introduction of symmetry scores for

immediate feedback to the athlete on a

smartphone/tablet.

Stamm, A. and Shlyonsky, I.

Freestyle Swimming Analysis of Symmetry and Velocities using a MEMS based IMU: Introducing a Symmetry Score.

DOI: 10.5220/0010134700380043

In Proceedings of the 8th International Conference on Sport Sciences Research and Technology Support (icSPORTS 2020), pages 38-43

ISBN: 978-989-758-481-7

2 METHODS

2.1 Instrumentation

This study used a MetaMotionC 9-axis IMU with 3-

axis accelerometer, 3-axis gyroscope, 3-axis

magnetometer, barometer, and light sensor. It comes

with built in Bluetooth for real time streaming and

communication, internal memory for data storage, a

CR2032 battery which last up to 48 hours of

recording, a weight of less than 7g, and physical

dimensions of 25 mm x 25 mm x 4 mm (width, height,

depth) in a standard casing.

Figure 1 shows the sensor placed at the swimmer. The

IMU was set at 100 Hz sampling rate.



Figure 1: Sensor position and orientation at the swimmer.

2.2 Data Collection

Ten swimmers (9 males, 1 female, 37.5±12.4 years,

179.4±6.5 cm, 76.4±11.7 kg, see Table 1) with

different levels of experience took part in this study.

The experiments have been carried out in line with

the Helsinki protocol for human research.

Data were collected at a 25m temperature controlled

indoor pool. The swimmers where asked to perform

an individual warm-up procedure to reduce the risk of

injuries before they swam four laps which were

recorded at different efforts. We expected a large

variability as the efforts have been self-determined by

the swimmers (low, medium, and full).

In this study, the IMU was taped firmly to the lower

back of the swimmer to reduce unwanted IMU

Table 1: List of swimmers with their height, mass, age and

experience.

Swimmer Height

(cm)

Mass

(kg)

Age

(years)

Experience Gender

1 183 79 29 international male

2 168 53 49 amateur female

3 172 82 37 amateur male

4 182 82 48 intermediate male

5 192 82 23 amateur male

6 178 81 52 amateur male

7 183 79 29 international male

8 183 85 26 national male

9 175 80 44 amateur male

10 180 90 51 amateur male

movements and to minimize skin movements. The

forward direction is represented by the a

, the

mediolateral direction by a

, and the anterior-

posterior direction by a

The data were downloaded via Bluetooth at the end

of each training session using the App “MetaBase”

provided by Mbientlab on an Android device. The

downloaded data were then further send via email to

the analysis team.

2.3 Data Processing

Data processing was automatically undertaken using

multiple Python scripts which were programmed to

find important parameters to athletes and coaches.

The acceleration data recorded by the IMU and sent

via email to the analysis team was firstly converted to

gravitational units before it was high-pass filtered

with a cut-off frequency of 0.3 Hz to seperate the

sensor orientation from the wanted acceleration

signal (James & Wixted, 2011; Stamm, James, &

Thiel, 2012; Stamm & Thiel, 2015; Stamm, Thiel,

Burkett, & James, 2011). This filter was applied to

remove the gravity signal form the acceleration

signal. A zero crossing algorithm was further applied

to the data to automatically separate the left and right

arm strokes.

Figure 2 shows the recorded acceleration signal for

one swimmer (blue) with the present sensor

orientation component (red) which was removed

from the recorded signal before it was further

processed. The zero crossing algorithm was then

applied to the gravity corrected mediolateral

acceleration data (body-roll) to find the individual left

and right arm strokes (see Figure 3).

Freestyle Swimming Analysis of Symmetry and Velocities using a MEMS based IMU: Introducing a Symmetry Score



Figure 2: Raw acceleration (blue) with the overlapping

gravity component (red).

Figure 3: Gravity corrected body-roll acceleration signal

(blue) with the overlapping zero-crossing detection result

(red).

A lap velocity and lap distance profile were

calculated as described by (Stamm et al., 2012) to

investigate the symmetry of the swimmer in terms of

timing, max/min arm velocity, distance, stroke rate,

and stroke length.

Figure 4 presents a typical lap velocity profile for a

25m freestyle swimming lap with the push-off phase,

swimming phase, and the stop phase. The focus for

the automatic symmetry detection was hereby set on

the swimming phase. The algorithm was set to detect

the second stroke at the start, as sometimes the first

stroke coincides with the push-off phase which leads

to a slightly disturbed acceleration signal, and to stop

at the second last stroke, as the last stroke quite often

coincides with the stop of the lap.



Figure 4: Lap velocity profile of a 25m freestyle swimming

lap.

Figure 5 shows a typical left and right arm intra-

stroke velocity profile extracted from a 50m freestyle

swimming phase whereby Figure 6 presents the left

and right arm distance extracted from the lap distance

profile after the zero-crossing detection algorithm

was applied.



Figure 5: Left (a) and right (b) arm intra-stroke velocities of

the swimming phase of a 25m freestyle swimming lap.



Figure 6: Left (a) and right (b) arm distances of the

swimming phase of a 50m freestyle swimming lap.

The investigated symmetries are now translated into

a simple symmetry score for the three individual

investigated parameters (stroke duration, length,

velocity) so that the swimmer can directly interpret

the score to help improving the swimming style (see

equation 1-3).

It needs to be mentioned that the symmetry scores

have been calculated using average left and right arm

timings, lengths, and velocities.

mmert

score

= 1 – (t

left

– t

) (1

)

Equation 1 describes the symmetry score for the

stroke duration considering the difference between

the left and right arm stroke. An ideal symmetry

would therefore always provide the result t=1, while

t<1 would present longer left arm stroke duration, and

t>1 would present a longer right arm stroke duration.

mmert

score

= 1 – (l

left

– l

) (2

)

Equation 2 describes the symmetry score for the

length of the stroke considering the difference

between the left and right arm stroke. An ideal

symmetry would be described by l=1, while l<1

would describe a longer left arm distance and l>1

would describe a longer right arm distance.

mmert

score

= 1 – (v

left

– v

) (3

)

0 5 10 15 20 25 30 35 40 45

-1

(g)

0 5 10 15 20 25 30 35 40 45

-1

(g)

0 5 10 15 20 25 30 35 40 45

(g)

0 5 10 15 20 25 30 35 40 45

Time (s)

0.5

1.5

2.5

tot

(g)

0 5 10 15 20 25 30

Time (s)

icSPORTS 2020 - 8th International Conference on Sport Sciences Research and Technology Support

Table 2: Left and right arm mean stroke durations, stroke lengths, and average velocities including the standard deviation

(SD) for the first lap of each swimmer.

Swimmer LEFT Arm Right Arm Left Arm Right Arm Left Arm Right Arm

Stroke Duration

(s) ± SD

Stroke Duration

(s) ± SD

Length (m)

± SD

Length (m)

± SD

Velocity

(m/s) ± SD

Velocity (m/s) ± SD

1 1.21 ± 0.11 0.98 ± 0.03 2.41 ± 0.31 1.85 ± 0.09 1.96 ± 0.18 1.89 ± 0.24

2 1.07 ± 0.13 1.12 ± 0.10 0.83 ± 0.10 0.88 ± 0.08 0.79 ± 0.07 0.79 ± 0.11

3 0.82 ± 0.07 0.83 ± 0.07 0.83 ± 0.07 0.89 ± 0.11 1.02 ± 0.13 1.06 ± 0.19

4 0.95 ± 0.10 0.99 ± 0.06 1.19 ± 0.12 1.28 ± 0.07 1.26 ± 0.19 1.29 ± 0.20

5 1.01 ± 0.09 0.86 ± 0.05 1.08 ± 0.11 0.92 ± 0.07 1.07 ± 0.10 1.06 ± 0.12

6 0.84 ± 0.04 0.96 ± 0.06 0.94 ± 0.05 1.08 ± 0.06 1.13 ± 0.24 1.12 ± 0.23

7 1.32 ± 0.18 1.51 ± 0.29 1.21 ± 0.21 1.31 ± 0.21 0.91 ± 0.15 0.87 ± 0.18

8 1.11 ± 0.07 1.07 ± 0.04 1.39 ± 0.10 1.23 ± 0.04 1.26 ± 0.14 1.17 ± 0.11

9 0.90 ± 0.09 0.93 ± 0.09 0.80 ± 0.08 0.80 ± 0.08 0.88 ± 0.12 0.87 ± 0.10

10 1.38 ± 0.06 1.36 ± 0.04 1.21 ± 0.07 1.14 ± 0.06 0.88 ± 0.13 0.83 ± 0.11

Equation 3 describes the symmetry score for the

average stroke velocity considering the difference

between the left and the right arm. An ideal symmetry

would be described by v=1, while v<1 would describe

a faster left arm velocity and v>1 would describe a

faster right arm velocity.

It should be noted that larger deviations come from

the fact that the symmetry scores were not

normalised.

All automatic calculations have been validated

against manual data analysis undertaken in MATLAB

as described by (Stamm et al., 2012). There has been

no significant difference between the automatic and

manual data analysis found.

3 RESULTS

A total of 30 freestyle swimming laps have been

analyzed for the participants (see

Table 1).

The zero-crossing detection algorithm applied to the

acceleration, lap velocity, and lap distance profile

data allowed the separation of left- and right arm

strokes, thus allowed the symmetry investigation of

the involved swimmers with the results of the first

swim of each swimmer being presented in Table 2.

The symmetry scores were calculated according to

equations 1-3 which have been described in the

previous chapter and are presented in Table 3. It can

be seen that the symmetry scores for swimmers who

had a large difference between the left and right arm

duration/length/velocity are significantly below or

above the ideal score of 1.

Table 3: Symmetry scores for the first lap of each swimmer

calculated under usage of mean stroke time, length, and

velocity

Swimmer

Time

symmetry

score

Length

symmetry

score

Velocity

symmetry

score

1 0.77 0.44 0.93

2 1.05 1.05 1.00

3 1.02 1.06 1.05

4 1.04 1.09 1.04

5 0.85 0.84 1.00

6 1.12 1.14 1.00

7 1.19 1.10 0.97

8 0.96 0.84 0.91

9 1.02 1.00 0.99

10 0.98 0.93 0.96

Whereby a score smaller 1 for the stroke duration

means that the left arm took longer, a score smaller 1

for the length means that the left arm traveled a longer

distance, and a score smaller 1 for the average

velocity meant that the left arm had reached a larger

velocity.

For the purpose of swimming methodology more

attention should be given to the situations where all

three scores are either smaller or larger than 1.

4 CONCLUSIONS

The arm symmetry of ten swimmers performing

freestyle swimming laps with different levels of

experience has been investigated using a MEMS

based IMU. The recorded tri-axial acceleration signal

was firstly high-pass filtered to remove the unwanted

gravity component of the acceleration signal. The

gravity corrected acceleration signal was then used to

Freestyle Swimming Analysis of Symmetry and Velocities using a MEMS based IMU: Introducing a Symmetry Score

calculate the lap velocity profile using an

approximation to the numerical integration as well as

the lap distance profile (Stamm et al., 2012). A zero-

crossing detection algorithm was then applied to

separate the left from the right arm strokes to facilitate

the arm symmetry investigations. The results found

average left and right arm velocities in the range from

0.53 m/s up to 2.08 m/s which goes in line with (Craig

& Pendergast, 1979; Craig et al., 2006; Stamm et al.,

2011).

It was further proposed to introduce three simple

numbers to provide (amateur) swimmers with an

index to show them potential improvements in their

applied swimming style. The three symmetry scores

have been based on the automatic analysis of freestyle

swimming laps and the parameters: stroke timings;

stroke length; and average stroke velocity. The

simplicity of the proposed symmetry scores reflects

in a simple way of interpretation as a perfect

symmetry would be reflected by the score 1. A

smaller score meant that the left arm took longer in

terms of arm timing; that the left arm had a longer

distance, and that the left arm had reached a higher

velocity, respectively. A score larger 1 reflects the

symmetry shift towards the right arm.

Considering the simplicity of the proposed symmetry

scores, it can be understood by every swimmer and

directly translated to a change in the swimming style

to further improve the technique.

First feedback was sought by the authors from

swimmers participated in that study and proofed that

this simple symmetry numbers were widely accepted

by the participants as an easy and understandable

symmetry score.

It is evident that the proposed methods present simple

symmetry scores and that they can be used to help

swimmers improving their swimming style. It can be

concluded that this simple method can be used to

substitute more complex equipment and therefore

help the swimmer to easily improve their swimming

symmetry.

The next step would be to involve coaches to be able

to have their feedback to further improve the methods

applied and optimise the form of presentation. This

could be i.e. to provide the swimmers with a simple

to use app to which the IMU can be connected to

start/stop the recording; upload the data for an

automatic analysis; and a graphical presentation of

the results, and individual stroke analysis. All of this

should be made available to the swimmer within

minutes to be able to change the swimming style

while still performing the training session.

ACKNOWLEDGEMENTS

The authors would like to thank all swimmers who

participated in this study.

REFERENCES

Callaway, A. J. (2015). Measuring Kinematic Variables in

Front Crawl Swimming Using Accelerometers: A

Validation Study. Sensors, 15(5).

doi:10.3390/s150511363

Craig, A. B. J., & Pendergast, D. R. (1979). Relationships

of stroke rate, distance per stroke, and velocity in

competitive swimming. Medicine & Science in Sports

& Exercise, 11(3), 278-283. Retrieved from

http://journals.lww.com/acsm-

msse/Fulltext/1979/01130/Relationships_of_stroke_rat

e,_distance_per_stroke,.11.aspx

Craig, A. B. J., Termin, B., & Pendergast, D. R. (2006).

Simultaneous recordings of velocity and video during

swimming. Portugese Journal of Sport Sciences, 6, 32-

36.

Dadashi, F., Millet, G., & Aminian, K. (2013). Inertial

measurement unit and biomechanical analysis of

swimming: an update. 61, 21-26. Retrieved from

http://infoscience.epfl.ch/record/189819/files/Sportme

dizin%20-%20Dadashi%20et%20al-%202013.pdf

Guignard, B., Rouard, A., Chollet, D., & Seifert, L. (2017).

Behavioral Dynamics in Swimming: The Appropriate

Use of Inertial Measurement Units. Frontiers in

Psychology, 8, 383. Retrieved from

https://www.frontiersin.org/articles/10.3389/fpsyg.201

7.00383

James, D. A., & Wixted, A. (2011). ADAT: A Matlab

toolbox for handling time series athlete performance

data. Procedia Engineering, 13, 451-456.

doi:10.1016/j.proeng.2011.05.113

Magalhaes, F. A. d., Vannozzi, G., Gatta, G., & Fantozzi,

S. (2015). Wearable inertial sensors in swimming

motion analysis: a systematic review. Journal of Sports

Sciences, 33(7), 732-745.

doi:10.1080/02640414.2014.962574

mbientlab. (2020). MetaMotionC (MMC). Retrieved from

https://mbientlab.com/metamotionc/

Stamm, A., James, D., & Thiel, D. (2012). Velocity

profiling using inertial sensors for freestyle swimming.

Sports Engineering, 1-11. doi:10.1007/s12283-012-

0107-6

Stamm, A., Thiel, D., Burkett, B. J., & James, D. A. (2009).

Roadmapping performance enhancement measures and

technology in swimming. The Impact of Technology in

Sport II, 213-217. Retrieved from

icSPORTS 2020 - 8th International Conference on Sport Sciences Research and Technology Support

http://research.usc.edu.au/vital/access/manager/Reposi

tory/usc:6000

Stamm, A., & Thiel, D. V. (2015). Investigating Forward

Velocity and Symmetry in Freestyle Swimming Using

Inertial Sensors. Procedia Engineering, 112, 522-527.

doi:http://dx.doi.org/10.1016/j.proeng.2015.07.236

Stamm, A., Thiel, D. V., Burkett, B., & James, D. A.

(2011). Towards determining absolute velocity of

freestyle swimming using 3-axis accelerometers.

Procedia Engineering, 13, 120-125.

doi:10.1016/j.proeng.2011.05.061

Freestyle Swimming Analysis of Symmetry and Velocities using a MEMS based IMU: Introducing a Symmetry Score