Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 72 (2014) 132 – 137

The 2014 conference of the International Sports Engineering Association

Estimating energy expenditure during front crawl

swimming using accelerometers
Nikolai B. Nordsborga, Hugo G. Espinosab, David V. Thielb,*
a
Department of Exercise and Sports Sciences, University of Copenhagen, Copenhagen 2100, Denmark
b
Griffith School of Engineering, Griffith University, Brisbane QLD 4111, Australia

Abstract

The determination of energy expenditure is of major interest in training load and performance assessment. Small, wireless
accelerometer units have the potential to characterise energy expenditure during swimming. The correlation between absorbed
oxygen versus flume swimming speed and absorbed oxygen versus the three axis acceleration recorded on the sacrum, wrist
and ankle for swimmers of varying abilities was calculated using Bland-Altman analysis of variance through parallel regression
lines fitted for 60 participants, who swam at three different speeds for 6 min duration with 2 min rest times. Swimmers showed
a strong positive relationship between VO 2 and RMS acceleration on the wrist (r = 0.77) and ankle (r = 0.73) sensors but not on
the sacrum (r = 0.46). The sacrum data was split into elite and novice swimmers, resulting in a strong correlation for elite
swimmers and a poor correlation for novice swimmers.
A robust biomechanical technique for the determination of the energy expenditure of swimmers of different categories and
genders from acceleration data has been developed.
Crown Copyright © 2014 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
© 2014 The Authors. Published by Elsevier Ltd.
Selection and peer-review under responsibility of the Centre for Sports Engineering Research, Sheffield Hallam University
Selection and peer-review under responsibility of the Centre for Sports Engineering Research, Sheffield Hallam University.

Keywords: Swimming; energy expenditure; accelerometer; velocity; flume; ANOVA

* Corresponding author. Tel.: +61-7-3735-7192; fax: +61-7-3735-5198.

E-mail address: d.thiel@griffith.edu.au

1877-7058 Crown Copyright © 2014 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
Selection and peer-review under responsibility of the Centre for Sports Engineering Research, Sheffield Hallam University
doi:10.1016/j.proeng.2014.06.024
 Nikolai B. Nordsborg et al. / Procedia Engineering 72 (2014) 132 – 137 133

1. Introduction

Determining workload during swimming is of importance in relation to recreational swimming in a health

perspective and for competitive swimmers in order to quantify training load. Energy expenditure during swimming
can be determined directly by measurement of oxygen uptake: Rodriguez F.A. et al. (2000) or indirectly by use of
doubly labelled water and heart rate registration: Livingstone M.B.E. et al. (2003). Of these technologies, only
heart rate monitors are possible to use on a regular basis in the wet environment. However, there are several
drawbacks in using heart rate monitors that have led investigators to try to estimate training load by use of
subjective reports of rate of perceived exertion. While swimmers show a high compliance with the prescribed
training programme, there is a poor relationship between the coaches and swimmers interpretation of swimming
intensity: Stewart A.M. et al. (1997). In addition to the practical problems in using heart rate monitors, it is a
problem to quantify the energy expenditure of work performed above the speed that is expected to elicit maximal
oxygen uptake, as in for example during interval based training. Thus, new technologies which can precisely
determine the energy expenditure during swimming at all speeds are warranted.
For some years, accelerometers have been used to monitor human movement and relate this to the energy
expenditure of athletes and the tasks associated with every-day living: Chen K.Y. et al. (1997). More recently
preliminary work on the application of accelerometers in monitoring energy expenditure in swimming and on the
velocity profiling in freestyle swimming using a purpose-built inertial sensor has been reported: Stamm A. et al.
(2012).
Accelerometers can provide accurate timing information for swimmers including stroke rate and lap times, and
more recently, the calculation of glide velocity: Stamm A. et al. (2013). The stroke and lap times are determined
from features in the acceleration profile: Hagem R.M. et al. (2013). The velocity is determined through filtering
and integration of the acceleration data. However, it is unknown if energy expenditure during swimming can be
determined by accelerometers. Thus, the aim of the present study is to investigate if accelerations measured during
swimming can be related to the energy expenditure determined by direct measurement of oxygen uptake in
swimmers of different skill levels (elite, sub-elite, competitive and amateur swimmers). The correlation was
calculated using Bland-Altman analysis of variance through parallel regression lines fitted for all participants.

2. Methods

2.1. Swimmer selection

Sixty swimmers (fifteen females and forty-five males) were enrolled in this study. All were required to
undertake measurements of weight and height, and to self report age, swimming ability (when known, the 100 m
and 400 m times), prior injuries and current weekly exercise regime. The swimming ability was self-ranked as non-
swimmer (1), novice (2), recreational (3), competitive (4) and national/elite (5). The swimmers were informed of
the reasons for the study and signed a participation consent form. The study was approved by the Griffith
University Ethics Committee as project ENG/05/10/HREC.

2.2. Swim details

The swimmer was fitted with the accelerometers using medical tape and asked to both stand vertically and then
lie in the horizontal glide position for 5 s to calibrate the accelerometers. The swimmer was fitted with the face
mask and commenced the swim trial (Figure 1). Three 6 min swims were interspersed with 2 to 5 min rest periods.

2.3. Acceleration

The acceleration a i (t) at time t along the ith axis (i = x, y, and z) is given by the equation

2
ai  t  g cos T  t    i ˜ a  t   Z  t  i ˜ r  t  (1)
134 Nikolai B. Nordsborg et al. / Procedia Engineering 72 (2014) 132 – 137

where g is the gravitational acceleration, T(t) is the time dependent angle between the axis and the vertical
direction, i is the unit vector, a(t) is the linear acceleration, and Z(t) is the angular velocity of rotation at a distance
r(t) from the centre of rotation. The calculation of swim velocity and distance travelled have been demonstrated
using accelerometer measurements despite the error and uncertainties involved in the single and double
integrations required: Stamm A. et al. (2012).
As energy expenditure can only be related to the dynamic components of acceleration, the gravitation
acceleration must be removed. This includes the weight of the feet. For the three orthogonal axes j = x, y, and z, the
root mean acceleration a j,rms was calculated using a time window of fixed sample length

n
1 2 2
(2)
a j ,rms
n
¦a
i 1
j ,i  aj   aj 

where a j,i is the ith time sample in the j direction, n is the number of samples in the window, and the mean value in
the time window is

¦a j ,i
aj i 1 (3)
n

To reduce window induced fluctuations, the length of the time window was chosen to ensure that more than six
strokes were included in the window at the slowest speed. The total acceleration for each unit A tot was calculated
by summing the three RMS acceleration values as

3
ATOT ¦a 2 (4)
j , rms
j 1

Equations 1 and 4 were applied to the complete acceleration data set for every swimmer. The steady state RMS
acceleration was noted towards the end of every 6 min swim.

2.4. Instrumentation

Following the techniques used in previous swimming investigations with logging accelerometers: Stamm A. et
al. (2012), three triaxial accelerometer units: Davey N., et al. (2008), were taped to the swimmer’s sacrum, left
wrist and left ankle (Figure 1). By placing one unit as close as possible to the centre of mass, the dynamic effect of
body roll on the acceleration measurements is minimised (r < 40 mm), however, the gravitational changes will be
recorded given the window selected for equation 2. Thus a x,rms is a measure of body roll. The wrist coordinate
system is different to that of Ohgi (2002) who noted maximum acceleration during the in sweep phase and the
upsweep and recovery phase. The accelerometer was set to record data at 100 Hz. The RMS window was set to 6 s
(n = 600). The total flume time required for each swimmer was approximately 30 min. Figure 1 shows the
swimming flume, the oxygen consumption unit (Ox) and three accelerometer units located at the left wrist (n1),
sacrum (n2) and left ankle (n3), with the orientation of the sensor axes with respect to the human body. In the
swimming position, the swimmer moves principally in the y direction for the sacrum unit and the x and z directions
for this unit are the mediolateral and the vertical directions respectively.
In the flume, swimmers wore a face mask and a nose plug to ensure the gas analyser recorded the oxygen intake
and exhaled breath: Rodriguez F.A. et al. (2008). Oxygen consumption (VO 2 ) was recorded by a breath-by-breath
system adapted for swimming (Quark B2, Cosmed). The steady state VO 2 was taken as the mean value of 41
breaths at the end of each 6 min. The velocity of the water in the flume was controlled and recorded using a
calibrated propeller driven sensor. The swimmer was manually timed for 6 min or until the oxygen consumption
 Nikolai B. Nordsborg et al. / Procedia Engineering 72 (2014) 132 – 137 135

had reached a steady state for at least 30 s.

Fig. 1. Swimming flume, oxygen consumption unit (Ox) and three accelerometer units located at the left wrist (n1), sacrum (n2) and left ankle
(n3), with the orientation of the sensor axes with respect to the human body.

3. Results

3.1. Energy expenditure and swim speed

The relation between the oxygen consumption (VO 2 ) and the flume velocity for each swimmer with three swim
speeds was determined through multiple regression analysis: Bland and Altman (1995). The analysis of variance
for the regression is described in Table 1, and shows how the variability in VO 2 can be partitioned into
components due to different sources.

Table 1. Analysis of variance between the oxygen consumption (VO2) and the flume velocity for each
of the 60 swimmers (15 females and 45 males).
Source of Degree of Sum of Mean Variance Probability
variation freedom squares square ratio (F)
Swimmers 46 5.9431e7 1.2919e6 13.97 0
Velocity 1 1.0584e7 1.0584e7 114.48 0
Residual 55 8.598e6 9.2451e4

The magnitude of the correlation coefficient r within swimmers is calculated as (Bland and Altman, 1995)

Sq 1.0584 u107 (5)

r 0.75
S q  Res 1.0584 u107  8.598 u106

where S q and Res are the sum of squares and the residual sum of squares, respectively, for the flume velocity from
Table 1. A strong positive correlation r = 0.75 shows that an increase in the velocity for an individual swimmer
was associated with an increased VO 2 , regardless of the differences between the swimmers.
136 Nikolai B. Nordsborg et al. / Procedia Engineering 72 (2014) 132 – 137

3.2. Swim speed and total acceleration

An analysis of variance (ANOVA) of the total RMS acceleration at the sacrum, the wrist and the ankle, as a
function of energy expenditure (measured as VO 2 ) was implemented for all swimmers with three swim speeds for
each one. Figures 2a and 2b show the ANOVA through parallel regression lines at the wrist and at the ankle. The
magnitude of the correlation coefficients using (5) at the wrist is r = 0.77, at the ankle is r = 0.73, and at the sacrum
is r = 0.46.

5000 5000

4500 4500

4000 4000

3500 3500
VO2 (ml/min)

VO2 (ml/min)
3000 3000

2500 2500

2000 2000

1500 1500

1000 1000

500 500

0 0
8 10 12 14 16 18 20 22 24 10 15 20 25 30 35
RMS Wrist Atot (m/s2) RMS Ankle Atot (m/s2)
(a) (b)

Fig. 2. Analysis of variance through parallel regression lines of the total RMS acceleration, as a function of energy expenditure (measured as
VO 2 ). (a) at the wrist and (b) at the ankle.

Since the sacrum acceleration data was found to be poorly correlated, the data was split into elite and novice
swimmers, in order to identify the trend of the parallel regression lines, defined as having a 100 m swim speed of
greater than 1.6 m/s for elite and less than 1.3 m/s for novice. Although the wrist and ankle positions had a strong
positive correlation, the elite swimmers showed higher correlation than novice swimmers in both positions. The
correlation coefficients and probabilities at the three locations are given in Table 2.

Table 2. Magnitude of the correlation coefficients and probability at the sacrum, wrist and ankle
for elite and novice swimmers with three swim speeds for each one.
Correlation coefficient (r) Probability
Participant Elite Novice Elite Novice
Sacrum 0.9036 0.3722 < 0.0008 < 0.0233
Wrist 0.9614 0.8252 < 0.0001 < 0.0001
Ankle 0.9414 0.6465 < 0.0001 < 0.0002

A summary of the mean and standard deviation for all acceleration data is given in Table 3. The data from three
swims for each swimmer was used to calculate the tabulated mean and standard deviation. All units in m/s2.

Table 3. Mean and standard deviation for total acceleration (Atot) components at three body locations
(sacrum, left wrist and left ankle) for males (n = 45) and females (n = 15). All units in m/s2.
Male (n = 45) Female (n = 15)
RMS acceleration component Mean STD Mean STD
Sacrum (Atot) 1.25 0.17 1.22 0.24
Wrist (Atot) 1.78 0.29 1.70 0.23
Ankle (Atot) 2.09 0.56 1.78 0.49
 Nikolai B. Nordsborg et al. / Procedia Engineering 72 (2014) 132 – 137 137

4. Conclusions

An investigation of the relation between accelerations measured during swimming and the energy expenditure
determined by direct measurement of oxygen uptake in swimmers of different skill levels, has been presented in
this paper. Triaxial accelerometer sensors placed in the swimmer’s sacrum, left wrist and left ankle were used to
record the acceleration of three 6 min swims of each subject.
A strong positive correlation r = 0.75 showed that an increase in the velocity for an individual swimmer was
associated with an increased VO 2 . Also, a strong positive correlation coefficient was observed for the wrist and
ankle data, however, the sacrum data was poorly correlated. It was necessary to split the data between elite and
novice swimmers to establish the relationship for elite swimmers. The total RMS ankle acceleration was larger
than the wrist and the sacrum measurements. This occurs because the radius of rotation is largest for the ankle. The
y component of the sacrum data (direction of the swim) was found to be the smallest with a mean of 0.23. Craig
and Pendergast (1979) reported that the variation in the intra-stroke velocity during free style swimming was
+20%. A relatively small and gradual change in acceleration data substantiates this observation.
The determination of the energy expenditure of swimmers of different categories and genders from acceleration
data has been developed by an efficient biomechanical technique.

Acknowledgements

The authors gratefully acknowledge Ronnie Dalsgaard for his assistance with the recruitment of the swimmers
and the data collection at the flume and pool. The research was supported by a Queensland International
Fellowship covering Thiel’s three month visit to Copenhagen. Swimming research at Griffith University is funded
by the Australian Research Council and the Centre of Excellence in Sports Science at the Queensland Academy of
Sport. The authors are indebted to Sue Hooper for her support in the project.

References

Bland, J.M., Altman, D.G., 1995. Calculating correlation coefficients with repeated observations: Part 1 – correlation within subjects. BMJ,
310, 446.
Chen, K.Y., and Sun, M., 1997. Improving energy expenditure estimation by using a triaxial accelerometer, Journal of Applied Physiology 83,
2112-2122.
Craig, A.B., Pendergast, D.R., 1979. Relationships between stroke rate, distance per stroke, and velocity in competitive swimming. Medicine
and Science in Sport 11, 278-283.
Davey, N., Anderson, M., James, D.A., 2008. Validation trial of an accelerometer-based platform for swimming. Sports Technology 1, 202-
207.
Hagem, R.M., O’Keefe, S.G., Fickenscher, T., Thiel, D.V., 2013. Self contained adaptable optical wireless communications system for stroke
rate during swimming 13(8), 3144-3151.
Livingstone, M.B.E., Robson, P.J., Black, A.E., Coward, W.A., Wallace, J.M.W., Mckinley, M.C., Strain, J.J., McKenna, P.G., 2003. An
evaluation of the sensitivity and specificity of energy expenditure measured by heart rate and the Goldberg cut-off for energy intake: basal
metabolic rate for identifying mis-reporting of energy intake by adults and children: a retrospective analysis. European J. Clinical Nutrition,
57(3), 455-463.
Ohgi, Y., 2002. Microcomputer-based acceleration sensor device for sports biomechanics – stroke evaluation by using swimmer’s wrist
acceleration. Proceedings of IEEE Sensors Conference 1, 699-704.
Quark B2, Cosmed, Pulmonar Function Equipment. http://www.cosmed.com/
Rodriguez, F.A. 2000. Maximal oxygen uptake and cardiorespiratory response to maximal 400-m free swimming, running and cycling tests in
competitive swimmers. Journal of Sports Medicine and Physical Fitness 40(2), 87-95.
Rodriguez, F.A., Keskinen, K.L., 2008. Validity of a swimming snorkel for metabolic testing. International Journal of Sports Medicine 29, 120-
128.
Stamm, A., James, D.A., Burkett, B.B., Hagem, R.M., Thiel, D.V., 2013. Determining maximum push-off velocity in swimming using
accelerometers. Procedia Engineering 60, 201-207.
Stamm, A., James, D.A., Thiel, D.V., 2012. Velocity profiling using inertial sensors for freestyle swimming. Sports Engineering. DOI
10.1007/s12283-012-0107-6.
Stewart, A.M., Hopkins, W.G., 1997. Swimmers’ compliance with training prescription. Medicine and Science in Sports and Exercise 29(10),
1389-92.