Available online at www.sciencedirect.

com
Procedia
Engineering
Procedia Engineering
Procedia Engineering 2
00(2010) 2851–2856
(2009) 000–000

www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia

8th Conference of the International Sports Engineering Association (ISEA)

Visualization and motion analysis of swimming

Shinichiro Itoa* and Keisuke Okunob
a
Shinichiro Ito, 1-24-2 Nishi-sinnjuku, Shinjuku-ku, Tokyo, 163-8677, Japan
b
Keisuke Okuno, Waseda University, 2-579-15 Mikashima, Tokorozawa, Saitama 169-8050, Japan
Received 31 January 2010; revised 7 March 2010; accepted 21 March 2010

Abstract

It is an advantage to film a swimmer’s whole body in order to grasp the swimmer's technique. However, it is difficult to
understand details, such as twisting arms or stroke paths. Equipment was constructed which provides a continuous image of
swimming motions by blending underwater and overwater images with a video mixer. Underwater and overwater images were
taken by cameras loaded onto a cart placed on a poolside rail track. The view from above the swimmer was observed by an
overhanging camera to observe wave resistance.. In order to grasp detailed swimming motions, logger data were synchronized
with the actual motion images. Wavelet transform, a chronological frequency analysis, was also performed on these logger data
and the dominant frequency was viewed chronologically with different swimming styles.

c 2009
© 2010Published
Publishedby by Ltd. Ltd. Open access under CC BY-NC-ND license.
Elsevier
Elsevier

Keywords: observation; visualization; motion analysis; wavelet transform

1. Introduction

Swimming is the successive motion of propulsive and recovery movements in the water. Analysis of the
swimming motion can be provided to swimmers and coaches for improvement of performance. It is important to
understand the underwater motion in order to improve swimming technique. Tracking underwater cameras are often
used in competition but are not generally used in training because of their installation and cost. Morouço et al. [1]
and Soons et al. [2] were able to produce a continuous image of the swimming movements by blending underwater
and overwater images with a video mixer. One of the author, Ito [3], have also developed a same kind of devices and
can capture the swimmer's whole body motion including recovery overwater both for training and research. Both
underwater and overwater images were recorded by the cameras loaded onto a cart placed on a poolside rail track.
The view from above the swimmer including the waves generated around the swimmer could not be seen in this

* Corresponding author. Tel.: +81-90-8461-0487; fax: +81-3-3340-0108.

E-mail address: ito@cc.kogakuin.ac.jp

1877-7058 c 2010 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
doi:10.1016/j.proeng.2010.04.077
2852 S. Ito, K. Okuno / Procedia Engineering 2 (2010) 2851–2856
2 S.Ito & K. Okuno / Procedia Engineering 00 (2010) 000–000

recording. In order to compensate for this deficit, a view from above the swimmer was recorded, to consider wave
resistance.
It is still difficult to observe detailed movement, such as twisting arms underwater. In order to observe these
detailed motions of the forearm in swimming, Ohgi [4] captured 3D accelerations and 3D angular velocities by a
data logger and analyzed change in movement caused by fatigue. However, no actual motion images were used in
analyzing their data.
In this study, a whole body swimming image including a view from above, was taken and also synchronizing the
three-dimensional acceleration and the angular velocity data on the forearm and analyzed them to understand subtle
differences in S-shaped and I-shaped movements in free style. A chronological dominant frequency was obtained
by using wavelet transform in these obtained data.

2. Method

The experiments were performed in the 50m × 25m indoor swimming pool, Aqua Arena, at Waseda University.
The subject of the experiments was an elite swimmer of breast stroke who had won a gold medal in 4 × 100m
medley relay in Universiade Belgrade 2009.

2.1. Observation device

Figure 1 shows the outline of the form observation device in this study. It consists of an underwater cameras and
an overwater one in a frame installed on a cart placed on metal pipe railing with 38mm in diameter. A crane tripod
and a camera over the swimmer were installed on the same cart to capture the overhead images. The respective
images of the underwater and the overwater were synthesized as a whole body image by a video mixer. The images
taken over the swimmer were inserted into the video mixer image using a function of 4 image dividers. All of these
images were captured by 29.97Hz with normal video cameras. The cart can run with the swimmer at 2m/s.

2.2. Data logger

Figure 2 indicates a data logger of motion sensors which has 3 components of accelerometers, 3 components of
gyroscope, a depth sensor and propeller system speedometer built-in (PD3G3Gy made by Little Leonard company).
It is composed of a CPU, 256MB memory, 8 components of A/D converters in a cylindrical shape, has cylindrical
shape of 23mm in diameter and 235mm in length. The data logger can measure 8 hours continuously in 128Hz. The
data logger was set on the backside of the left forearm on a three-dimensional coordinate axis as shown in Fig. 3.

2.3. Data Analysis

Each data were sampled in 128Hz. A part of the data corresponding to swimming motion for the three-
dimensional acceleration and the angular velocity data taken from the data logger were extracted with Igor ProVer.6

Fig. 1. Outline of experimental equipment; Fig. 2. Data logger of motion sensor.

 S. Ito, K. Okuno / Procedia Engineering 2 (2010) 2851–2856 2853
S.Ito & K. Okuno / Procedia Engineering 00 (2010) 000–000 3

y-direction acceleration:
Arm extention
x-direction acceleration:
Arm side slide

z-direction acceleration:
Arm pull

Fig. 3. Definition of 3-components on data logger; Fig. 4. Morlet function as mother function of wavelet transform.

of the Wavemetric Co. Ltd. These wave data were synchronized with the swimming motion images by Pixel Runner
of Tellus Image Co. Ltd. by matching the motion image with the time at the arm’s entering the water and the peak
point of the data waves.
Regarding the logger signal processing, the wavelet transform was performed by using free macro software of
Ethographer (Sakamoto KQ [5]) on Igor Pro. The wavelet transformation is one of the techniques of the
chronological frequency analysis. The Morlet function of 15 cycles shown in Fig. 4 was employed as mother
wavelet in this study. A key advantage over Fourier transforms is temporal resolution. On the other hand wavelet
transforms capture both frequency and location information.

3. Results and Discussion

Figure 5(a) shows the composite picture of the swimming motion of underwater and overwater obtained by the
observation device. This image is captured by the underwater and overwater cameras loaded on a cart moving with
the swimmer. Top view through the overhung camera is shown in Fig. 5 (b). Generated wave around a swimmer is
observed clearly.
The composed image was viewed as if the swimmer in a water tank had been observed through plexiglas wall.
Figure 6 shows the composite underwater and overwater image of the swimming motion obtained by the observation
equipment. The picture shows as if the swimmer had been observed in a water tank through a window. Figure 6 also

Fig. 5. (a) Mixing picture of under and over water; (b) Overview at the same time.
2854 S. Ito, K. Okuno / Procedia Engineering 2 (2010) 2851–2856
4 S.Ito & K. Okuno / Procedia Engineering 00 (2010) 000–000

shows the three-dimensional accelerations and the angular velocity obtained by the data logger attached on the
forearm synchronized with the motion image. In the case of the breaststroke shown in Fig. 6, a sequence of opening
and closing motion of the arm appeared in Arm Slide in Fig. 6-1. Concerning the acceleration information of Arm
Extension in Fig.6-2, two peaks were seen in one stroke, although the motion of the arm itself was carried out
smoothly. Since the accelerometer has also gathered gravitational acceleration, the negative peak was obtained when
the arm was moving downward. The effect of gravitational acceleration was seen as well in the acceleration of Arm
Pull in Fig.6-3. On the other hand, three component angular velocity of Arm Pitch, Arm Roll, and Arm Yaw was
carrying out the complicated revolution shown in Fig.6-5,6,7 respectively.

2
[m/s ]] 1. Arm Slide Slide [deg/s] 5. Arm Pitch

2
[m/s ]] 2. Arm Extension
[deg/s] 6. Arm Roll

2 [deg/s] 7. Arm Yaw

[m/s ]] 3. Arm Pull

[m]] 4. Depth

Fig.6. Chronological 3D acceleration, 3D gyro and depth data of arm stroke synchronized with whole body image of under and over water

recovery phase
Time A Time B

Slice Section

stroke/pull

Fig. 7. Gyro sensor data of x-direction, wavelet transform for the same wave and transition of depth acquired with the sensor attached on
the forearm in the freestyle stroke.
 S. Ito, K. Okuno / Procedia Engineering 2 (2010) 2851–2856 2855
S.Ito & K. Okuno / Procedia Engineering 00 (2010) 000–000 5

Fig. 8(a) Section of TimeA; (b) Section of Time B

Figs. 8. Time sliced section data of wavelet transform output, Fig.4 at time A and B shown in Fig.4. The sliced section shows momentary
cycle (=1/frequency) analysis.

Fig. 9. Cycle sliced section data of wavelet transform output, Fig.7 at sliced section, Cycle 2.2 sec. This is a stroke cycle of freestroke in
this case. The bottom peak shows the turn phase.

In order to understand these data better, synchronizing this motion image and the data waves was a very effective
mean to grasp the differences in details of an individual swimmer's strokes even in the same swimming style.
The graph at the top of Fig. 7 shows the transition of flexion/extension motion of the forearm described with
angular velocity. The graph in the middle indicates the wavelet transform for the same data and the graph at the
bottom shows the transition of depth. Each of them was acquired with the sensor attached on the forearm in the
freestyle stroke. Compared the graphs at the top and the bottom, the angular velocity in the stroke phase is almost
constant, while that in the recovery phase decreased rapidly. The Fourier transform converts a spectrum space of a
certain period into a frequency space. On the other hand, The wavelet transform is the time series data of the
momentary frequency conversion in the same time space. In other words, the Fourier transform is less useful in
analyzing non-stationary data, where there is no repetition within the region sampled. The wavelet transform allows
the components of a non-stationary signal to be analyzed better. In the middle of Fig. 7, the thick striped pattern
appears in the dominant cycle for about 2 seconds of the respective movements. The output of wavelet transform
can be sliced in x and y direction. Figure 8(a) and 8(b) are the sections of Fig. 4 at Time A and B respectively.
These sliced sections show momentary cycle (=1/frequency) analyses. In these figures, the primary peak indicates
the stroke cycle. The secondary peak appears in the cycle of S-shaped stroke. Figure 9 shows cycle sliced section
data of wavelet transform output at Slice Section C in Fig. 7 in Cycle 2.2 sec. This is a stroke cycle of freestroke in
this case. The bottom peak shows the turn phase.
Thus, the wavelet transform converts motion waves to pattern designs. Three acceleration waves and three
angular velocity waves acquired with a motion logger indicate six motion patterns. Four swimming events, such as
freestroke, breaststroke, backstroke and butterfly stroke, consist of 24 pattern pictures. The difference of picture
2856 S. Ito, K. Okuno / Procedia Engineering 2 (2010) 2851–2856
6 S.Ito & K. Okuno / Procedia Engineering 00 (2010) 000–000

0.5 sec
1.0 sec
2.0 sec

Fig. 10 Gyro sensor data of x-direction, wavelet transform for the same wave and transition of depth acquired in the butterfly stroke.

patterns should lead to another motion analysis with wavelet transform. Figure 9 shows the transition of
flexion/extention motion of the forearm described with angular velocity, the wavelet transform for the same data and
the transition of depth in the butterfly stroke. Two stripe patterns appear in the wavelet transform picture. Each
stripe shows the basic stroke cycle and the cycle of a keyhole shaped stroke in the butterfly stroke.

4. Conclusion

The motion filming equipment was developed to show the whole body, a synthesized swimming image of the
underwater and overwater motion. Furthermore, this synthesized swimming image was synchronized with the data
obtained with the data logger which samples the 3D acceleration and 3D angular velocity of the forearm movement.
The wavelet transform was performed on the motion data and the various motions were visualised as a new
motion analysis method. As the implication of the wavelet transform, it should becomes possible to express the
difference in the strokes even in the same event or to express a collapse of the form by fatigue by showing the
motion as picture patterns.

References

[1] Soons, B., Colman V., Persyn U., Development of a multi-media system for kinesiological evaluation of swimming by experts in any pool,
In: Biomechanics and Medicine in Swimming X. Eds: Vilas-Boas, J.P., Alves, F. and Marques, A. Portuguese Journal of Sport Sciences 6(Supl
2) (2006), 277-279.
[2] Morouço P., Lima, A. B., Semblano, P., Fernandes, D., Gonçalves, P., Sousa, F., Fernandes, R., Barbosa, T., Correia, M. V., Vilas-Boas, J.
P., Validation of a cable speedometer for butterfly evaluation, In: Biomechanics and Medicine in Swimming X. Eds: Vilas-Boas, J.P., Alves, F.
and Marques, A. Portuguese Journal of Sport Sciences 6(Supl 2) (2006), 236-239.
[3] Ito, S, Swimming form analysis of competitive free style, Proc. Annual Meeting, Japan Society of Mechanical Engineers, No.03-1.(2003),
pp65-66.
[4] Ohgi, Y., Pattern matching application for the swimming stroke recognition, Biomechanics of Medicine in Swimming X, J.P.Vilas-Boas, F.
Alves, A. Marques, eds., Proc. of Xth International Symposium Biomechanics and Medicine in Swimming (2006), pp.69-70.
[5] Sakamoto KQ, Sato K, Ishizuka M, Watanuki Y, Takahashi A, Daunt F, Wanless S Can Ethograms Be Automatically Generated Using
Body Acceleration Data from Free-Ranging Birds? PLoS ONE 4(4) (2009): e5379.