This article was downloaded by: [University of Ottawa] On: 23 November 2014, At: 19:07 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Journal of Sports Sciences Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/rjsp20
Wearable inertial sensors in swimming motion analysis: a systematic review a
b
c
Fabricio Anicio de Magalhaes , Giuseppe Vannozzi , Giorgio Gatta & Silvia Fantozzi
ad
a
Department of Electrical, Electronic, and Information Engineering “Guglielmo Marconi”, University of Bologna, Bologna, Italy b
Interuniversity Centre of Bioengineering of the Human Neuromuscoloskeletal System, Department of Motor, Human and Health Sciences, Universita’ di Roma Foro Italico, Rome, Italy c
Department for Life Quality Studies, University of Bologna, Bologna, Italy
d
Health Sciences and Technologies – Interdepartmental Center for Industrial Research, University of Bologna, Bologna, Italy Published online: 30 Oct 2014.
To cite this article: Fabricio Anicio de Magalhaes, Giuseppe Vannozzi, Giorgio Gatta & Silvia Fantozzi (2014): Wearable inertial sensors in swimming motion analysis: a systematic review, Journal of Sports Sciences, DOI: 10.1080/02640414.2014.962574 To link to this article: http://dx.doi.org/10.1080/02640414.2014.962574
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Journal of Sports Sciences, 2014 http://dx.doi.org/10.1080/02640414.2014.962574
REVIEW ARTICLE
Wearable inertial sensors in swimming motion analysis: a systematic review
FABRICIO ANICIO DE MAGALHAES1, GIUSEPPE VANNOZZI2, GIORGIO GATTA3 & SILVIA FANTOZZI1,4 Department of Electrical, Electronic, and Information Engineering “Guglielmo Marconi”, University of Bologna, Bologna, Italy, 2Interuniversity Centre of Bioengineering of the Human Neuromuscoloskeletal System, Department of Motor, Human and Health Sciences, Universita’ di Roma Foro Italico, Rome, Italy, 3Department for Life Quality Studies, University of Bologna, Bologna, Italy and 4Health Sciences and Technologies – Interdepartmental Center for Industrial Research, University of Bologna, Bologna, Italy
Downloaded by [University of Ottawa] at 19:07 23 November 2014
1
(Accepted 1 September 2014)
Abstract The use of contemporary technology is widely recognised as a key tool for enhancing competitive performance in swimming. Video analysis is traditionally used by coaches to acquire reliable biomechanical data about swimming performance; however, this approach requires a huge computational effort, thus introducing a delay in providing quantitative information. Inertial and magnetic sensors, including accelerometers, gyroscopes and magnetometers, have been recently introduced to assess the biomechanics of swimming performance. Research in this field has attracted a great deal of interest in the last decade due to the gradual improvement of the performance of sensors and the decreasing cost of miniaturised wearable devices. With the aim of describing the state of the art of current developments in this area, a systematic review of the existing methods was performed using the following databases: PubMed, ISI Web of Knowledge, IEEE Xplore, Google Scholar, Scopus and Science Direct. Twenty-seven articles published in indexed journals and conference proceedings, focusing on the biomechanical analysis of swimming by means of inertial sensors were reviewed. The articles were categorised according to sensor’s specification, anatomical sites where the sensors were attached, experimental design and applications for the analysis of swimming performance. Results indicate that inertial sensors are reliable tools for swimming biomechanical analyses. Keywords: swimming, biomechanics, inertial sensors
Introduction Currently, minute differences in the race times of finalists and winners of swimming events have forced athletes, coaches and sport scientists to work harder to get a “fraction of a second” improvement (James, Davey, & Rice, 2004). The use of contemporary technology plays an important role in supporting performance enhancement from a biomechanical point of view by allowing detailed analysis of motion to be carried out and thus refine swimming technique (Pansiot, Lo, & Guang-Zhong, 2010). The analysis of in-water movements can improve the understanding of swimming biomechanics, enabling swimmers to perform closer to their full potential
(Callaway, Cobb, & Jones, 2009). Systematic biomechanical analysis identifies technical errors and monitors performance progress (Daukantas, Marozas, & Lukosevicius, 2008) of the swimming technique, which provides individuals with a more efficient swimming technique. Traditionally, motion measurements during swimming have been performed using video analysis method as gold standard (Ceseracciu et al., 2011; McCabe, Psycharakis, & Sanders, 2011; McCabe & Sanders, 2012; Vezos et al., 2007). Therefore, the use of underwater video cameras for recording, modelling and refining swimming strokes has become a common approach for elite swimmers (Callaway et al., 2009). This method requires a
Correspondence: Giuseppe Vannozzi, IUC-BOHNES, Department of Motor, Human and Health Sciences, Universita’ di Roma Foro Italico, Rome, Italy. E-mail: [email protected] © 2014 Taylor & Francis
Downloaded by [University of Ottawa] at 19:07 23 November 2014
2
Fabricio Ancio de Magalhaes et al.
tracking procedure for estimating the position of given points of interest (typically anatomical landmarks) using algorithms specifically developed to extract performance-related biomechanical variables (Ceccon et al., 2013; Pansiot et al., 2010). Marker tracking is the most accurate and comprehensible method for swimming motion analysis, which provides relevant data to coaches and athletes (Callaway et al., 2009); however, making this process fully automatic is difficult. Furthermore, the tracking of markers is time-consuming, error-prone due to water turbulence, and requires intensive data processing (Gourgoulis et al., 2008; Wilson et al., 1999). Moreover, a number of synchronised cameras are required to achieve a complete 3D kinematic description, increasing the costs of the procedure. Furthermore, video analysis can only be performed off-line, which delays the process of providing real-time feedback by the coach to the swimmer to address technical errors. An automatic tracking procedure was recently developed (Magalhaes et al., 2013; Olstad, Zinner, Haakonsen, Cabri, & Kjendlie, 2012) to minimise the time-consuming processing phase. Even though reflective markers and dedicated cameras were used (Olstad et al., 2012), concerns still existed for video-based motion analysis. Following the current trend of quantitative human movement analysis in different application contexts, inertial measurement units (IMUs) worn by the athletes have been proposed as an alternative tool for in-field sports performance analysis to overcome the limitations of video-based methods. Swimming biomechanics has been strongly oriented toward the use of IMU technology due to the above-mentioned difficulties in measuring the movement of swimmers in the water by means of common video-based methods. Wearable IMU sensors do not require a complex measurement setup in the swimming pool and are considered swimmer-centric (Pansiot et al., 2010). The current advent of micro-electro-mechanical systems technology provides practical IMU applications by reducing both the dimension and power consumption of the devices. This technology allows various setups and positioning on the swimmer’s body. Research in this field has provided evidence of the application of IMU sensors for monitoring human movement in sport. Furthermore, these sensors have meaningful advantages allowing reliable measurements in different environments, such as in the water, and providing specific environmental advantages that can improve data acquisition. Several authors have examined the movement patterns and sports techniques using accelerometers, gyroscopes and, eventually, magnetometers (Caruso, 1997; Jones, Jones, & Nenadic, 2013). Following the
emerging trends of this technology, swimming science first incorporated the use of accelerometers and, more recently, the gyroscopes; therefore, this review focused only on inertial sensors. For example, accelerometry data feedback given to athletes provided kinematic and performance consistency improvements in rowing (Anderson, Harrison, & Lyons, 2005). A wide range of areas were interested, including: ambulatory measurements, physical activity assessment, gait analysis, and evaluation of athletic performance (Callaway et al., 2009; Davey, Anderson, & James, 2008). Therefore, the use of IMUs has been shown to be an overall effective tool for monitoring human movement patterns, eventually including additional wearable devices such as pressure, temperature and humidity sensors. An increasing range of inertial sensors and protocols have been proposed for swimming performance assessment, and it is of interest to examine how they were used and how well they performed with respect to traditional movement analysis techniques. The increasing body of literature on this topic has not been reviewed. Therefore, the purpose of this article was to provide a systematic review regarding the current status of inertial and magnetic sensors for swimming performance assessment. The main objective of this review was to provide a framework to fully exploit the recent advances in miniaturised wearable technologies to obtain biomechanical data related to sport performance. Methods Review questions A systematic review of the literature regarding the inertial sensor-based performance assessment was carried out to address the following questions: (1) What are the existing inertial sensors applications for swimming motion analysis? (2) What type of inertial sensor was used? (3) Where were the inertial sensors attached? (4) How were the experiments conducted? (5) How was the performance evaluated using this approach? Article selection The research method was based on a systematic search for publications in the following scientific databases: PubMed, ISI Web of Knowledge (Science Citation Index Expanded), IEEE Xplore, Google Scholar, Scopus and Science Direct. The search included all journal and conference papers published before August 2013. These electronic search engines/databases were chosen as the most common databases related to methodological issues in applied biomechanics, covering most of the related literature in engineering, medicine and
Downloaded by [University of Ottawa] at 19:07 23 November 2014
Wearable inertial sensors in swimming biomechanics in sports. The searched keyword string was “(swimming OR swim OR swimmer) AND (inertial sensors OR inertial sensor OR accelerometer OR accelerometers OR acceleration OR accelerometry OR gyroscope OR gyroscopes OR gyroscopic OR magnetic sensor OR magnetometer OR IMU OR sensor fusion OR MEMS OR inertial tracking devices OR body worn sensors OR pervasive computing OR WSN OR INS OR WIMU OR BSN)” appearing in the title, abstract and keyword fields of the articles. Common acronyms used in the context of inertial sensors technology were included in the search string such as MEMS (Micro ElectroMechanical Systems), WSN (Wireless Sensor Network), INS (Inertial Navigation System), WIMU (Wireless Inertial Measurement Unit), BSN (Body Sensor Networks). Only English language articles were considered, and the initial total number of articles identified from all databases was 65 (Medline 28, ISI 10, IEEE Xplore 16, Google Scholar 12, Scopus 6, Science Direct 13), including journal articles, conference proceedings and book chapters. As an initial search strategy, both the title and abstract were first identified and read carefully to establish a first selection of articles. Relevance of the article was evaluated based on the pertinence to the topic “measurement of swimming biomechanics”. When it was not clearly indicated whether an article should be included or not, the complete article was retrieved from the local university libraries and fully reviewed by the authors. Considering also the deletion of unrelated and duplicated articles, 27 full papers were ultimately included in this review. The inclusion criteria were as follows: (1) the study dealt with human swimming motion analysis, and (2) the study involved accelerometers, gyroscopes and/or magnetic sensors as main measuring devices. The following parameters were extracted by the review team: the participants’ characteristics, anatomical region involved in measurement, study procedures, type of sensor/portability, biomechanical variables measured, data processing methods and statistical analyses, gold standard measurement system, primary outcomes of accuracy and/or reliability. Following this grid, all the 27 research articles were organised as shown in Table I.
Results Applications of inertial and magnetic sensing in swimming Developing and testing specific protocols to perform a 3D swimmers’ kinematic analysis using IMUs has been the aim of several authors. The validation of sensors has been widely carried out in running (Bergamini et al., 2012; Lee, Mellifont, & Burkett,
3
2010) and jumping (Picerno, Camomilla, & Capranica, 2011). However, only a few swimming validation trials exist (Dadashi, Crettenand, Millet, & Aminian, 2012; Davey et al., 2008; Stamm, James, & Thiel 2013). Most of the studies are focused case studies, as reported in Table I. Based on the information contained in Table I, the sport biomechanist can select the appropriate sensor (accelerometer or gyroscope or both) and the relevant variables to estimate. The first attempts in using inertial sensors in swimming referred to the phase analysis, a fundamental prerequisite for each biomechanical analysis in sports. The first study using IMUs in swimming was carried out by Ohgi, Yasumura, Ichikawa, and Miyaji (2000). They investigated the relationships between the wrist acceleration data and hand movement under the Maglischo’s stroke pattern definition (Maglischo, 1993). Examining the tri-axial wrist acceleration of front crawl strokes, they verified the presence of a repeatable pattern for each stroke style that led to a feasible stroke phase identification (entry, down sweep, out sweep and in sweep). In addition, Ohgi, Ichikawa, Homma, and Miyaji (2003) using the same approach discriminated the different phases of the breaststroke style (recovery, in-sweep and out-sweep). Furthermore, IMUs output could be used to discriminate among the different swimming styles. Slawson et al. (2008), Siirtola, Laurinen, Roning, and Kinnunen (2011) and Hou (2012) reported that the four competition swim styles could be extracted from accelerometers worn on the wrist or the upper-back. Pansiot et al. (2010), by deriving pitch and roll angles, were able to detect the type of stroke and the wall push-offs by means of a single miniaturised accelerometer-based sensor mounted on the swimmer’s goggles. A similar method was used by Vannozzi et al. (2010) in the analysis of the four swimming strokes using gyroscope data obtained using a single waist-mounted IMU. The characteristics of the swimmer’s stroke motion can be acquired using IMUs easily and continuously (Ohgi & Ichikawa, 2002; Ohgi, Ichikawa, & Miyaji, 2002) during the entire length of the pool, instead of focusing on a single stroke as typically allowed by traditional video analysis. Although several authors made attempts to insert IMUs in the swimming environment, only a few quantitative parameters were extracted from the registered signals, and a few processing methods were used for that purpose. Typically, the swim data analyses have been carried out with a personal computer using off-line algorithms after the swimming exercise. This represents a very useful procedure for specific technical analyses to be performed after the competition or the physical training; however, the long processing time needed typically does not allow the coaches to give their immediate
Beanland, Main, Aisbett, Gastin, and Netto (2013) Chakravorti, Le Sage, Slawson, Conway, and West (2013) Dadashi et al. (2012)
Dadashi, Crettenand, et al. (2013) Dadashi, Arami, et al. (2013) Davey et al. (2008) Fulton, Pyne, and Burkett (2009b)
Fulton, Pyne, and Burkett (2009a)
Fulton, Pyne, and Burkett (2011)
Hagem, Thiel, O’Keefe, and Fickenscher (2013) Hou (2012)
James et al. (2011)
Le Sage et al. (2010b)
2
5
9
10
11
13
14
12
7 8
6
4
3
Bächlin and Tröster (2012)
Source
1
Serial nos
1
3
2
2
1
4
1 1
2
3
1
1
1
4
Units
3D acc 2D gyro
3D acc; temperature, humidity and pressure sensors 3D acc 3D gyro
3D acc
3D acc 1D gyro
3D acc 1D gyro
3D acc 3D gyro 3D acc 3D gyro 3D acc 3D acc 1D gyro
3D acc 3D gyro
3D acc
3D acc
3D acc
Type
Not described
Not described
52 × 33 × 11 mm3
Not described
18 g
Not described
50 Hz
100 Hz
Sealing
Participant
MiniTraquaTM, Version 5, Cooperative Research Centre for Micro-technology, Australian Institute of Sport Not described
Plastic casing
Cited: James et al. (2004) Plastic casing
Not described
Hermetical sealing the plastic bags
Hermetical sealing the plastic bags
In-house custom-built waterproof package 1 swimmer (double membrane), known as “sealedfor-life solution” The prototype node was packaged to ensure 1 swimmer it was waterproof for the application (no further detail provided). (continued )
11 swimmers (male and females)
1 swimmer
7 elite and 19 recreational swimmers 7 trained swimmers (5 M, 2 F) 7 trained swimmers (4 M, 3 F) 6 elite swimmers 14 Paralympic swimmers (8 M, 6 F) 12 Paralympic swimmers (8 M, 4 F) 12 Paralympic swimmers (9 M, 3 F)
Water-tightness: plastic foil. Wrist: inside a 18 swimmers transparent plastic tube fixed using velcro (7 competitive, 8 fasteners. recreational and 3 occasional) Not described 21 swimmers (high level at least 5 sessions per week Not described 2 swimmers
Range between 5 Not described and 50 Hz
MiniTraquaTM, Version 5, Cooperative 100 Hz Research Centre for Micro-technology, Australian Institute of Sport 27 × 19 mm Not described Not described
100 Hz
52 × 33 × 11 mm3
100 Hz
500 Hz
500 Hz
Not described
100 Hz
256 Hz
Sample frequency
Not described 100 Hz
20.7 g
Not described
Not described
34 g
Weight
Cited: James et al. (2004) Cited: Fulton et al. (2009a)
Physilog®, BioAGM, CH
Physilog®, BioAGM, CH
Physilog®, BioAGM, CH
Not described
Not described
36 × 42 × 12 mm3
Sizes
Type and specification of sensors
Table I. List of the selected 27 articles organised with the indication of the first author and containing the different details related to sensor type and specification, number and type of participants, sensor location body area, analysed stroke type, spatiotemporal and biomechanical variables investigated, gold standard reference measurement, eventual data analysis methods and adopted statistics. Acronyms for the different styles are the following: front crawl (FC), breaststroke (BrS), backstroke (BaS) and butterfly (BF).
Downloaded by [University of Ottawa] at 19:07 23 November 2014
4 Fabricio Ancio de Magalhaes et al.
Le Sage et al. (2011b)
Le Sage et al. (2012)
Lee et al. (2010)
Lee, Ohgi, and James (2012)
Nakashima, Ohgi, Akiyama, and Kazami (2010) Ohgi et al. (2002)
Ohgi et al. (2003)
Pansiot et al. (2010) Siirtola et al. (2011)
Slawson et al. (2008)
Slawson, Justham, and Conway (2012) Stamm, Thiel, Burkett, and James (2011)
Stamm et al. (2013)
15
16
17
18
19
20
21
22 23
24
25
27
26
Source
Serial nos
Table I. (Continued).
1
1
1
1
1 2
1
1
1
1
1
1
1
Units
3D acc 3D gyro
3D acc 2D gyro 3D acc
3D acc 3D acc; temperature, humidity and pressure sensors 3D acc
2D acc
3D acc
3D acc 3D gyro
3D acc
3D acc 3D gyro
3D acc 2D gyro 3D acc
Type
190 Hz
100 Hz
200 Hz
Not described
25 Hz
Sample frequency
Not described
20 g
53 × 33 × 10 mm3
Not described
Not described
8g 18 g
62 g
100 Hz
100 Hz
Not described
50 Hz
50 Hz 5, 10, 25, 50 Hz
Not described
Prototype I: 50 g 128 Hz Prototype II: 78 g
Not described
Cited: Davey et al. (2008)
Not described
Not described
110 g
Weight
Not described
90 mm × 40 mm
Not described
Not described Not described
Not described, but smaller than 44 × 37 × 20 mm3
Prototype I: 88 × 21 mm Prototype II: 141.8 × 23.2 mm
Not described
Cited: Davey et al. (2008)
Not described
Not described
15 × 9 cm
Sizes
Type and specification of sensors
Sealing
1 swimmer
Participant
1 amateur level triathlete 1 male elite triathlete
(continued )
Not described but it can be seen in the figure 1 recreational 1 that the sensor was put in a plastic (or swimmer acrylic) housing. Waterproof casing 17 swimmers (8 junior elite and 9 retired elite)
Not described
Not described
2 top-level college swimmers (experiment 1) and 5 (4 male and 1 female) triathlon athletes (experiment 2) Not described but it can be seen in the figure 2 swimmers 1 that the sensor was put inside of a wristwatch. Rubber latex skin 1 swimmer Not described 11 amateurs and professionals
capsulated by aluminium alloy cylinder (200 m water resistant)
12 swimmers (8 males and 4 females) 6 swimmers (3 male and 3 female) 2 elite athletes (a swimmer and a triathlete) The wristwatch-style sensing unit is cased in 1 swimmer a waterproofed housing.
The sensor, battery and an antenna were packaged into a waterproof AquaPac Not described but it can be seen in the figure 4 that the sensor was packed in a plastic (or rubber) housing. Not described but it can be seen in the figure 1 that the sensor was housed in a plastic (or acrylic) cubic recipient. Cited: Davey et al. (2008)
Downloaded by [University of Ottawa] at 19:07 23 November 2014
Wearable inertial sensors in swimming 5
Upper back lower back right wrist
head
Lower back
Lower back (sacrum)
both forearm Lower back (sacrum)
Right arm, right leg
Lower back (sacrum)
dominant kicking leg
Both thighs both shanks
– – –
–
–
–
– –
–
–
–
– –
2
3
4
5
6
7
8
9
Body area
1
Serial nos
Table I. (Continued).
Only flutter kick
Only flutter kick
All
BrS
FC
FC
All
FC, BrS, BF
FC
Stroke type
– –
– – –
– – –
–
– –
–
– – –
– – – – – – –
–
– –
–
kick patterns kick-detection
kick count kick rate kick patterns
push-off stroke type stroke count metrics
–
–
–
Stroke phases (arm and – leg)
stroke phases inter-arm coordination
instantaneous velocity
stroke counts stroke rates lap counts
start (wall-push-off) – strokes end (wall-strike) Average velocity time per stroke (TPS) distance per stroke (DPS) stroke counts –
Variables
– –
–
–
– –
Video data (underwater)
Video data (underwater)
–
– –
–
– –
Hand timed and video data – (underwater) –
tethered apparatus – (SpeedRT®, ApLab, Rome, Italy). – – – Video data (2 underwater – cameras, 50 Hz; 189 – videos) Video data (2 underwater – cameras, 50 Hz; 168 – videos)
Video camera (2 fixed and 1 trolley mounted cameras; 25 Hz) Video data (1 underwater camera and 1 high speed video camera)
Normal stopwatch and real – accelerometer data –
Gold standard
Data analysis methods and statistics
(continued )
Phase detection algorithm (Hidden Markov Model) Mean absolute deviation, mean and standard deviation of the error; repeated measure ANOVA (difference not influenced by either of athlete or trial factors), sensitivity (93.5% arm phase, 94.4% leg phase) and specificity (96.2% arm phase, 97.2% leg phase) Maximum and minimum detection, threshold analysis, zero crossing algorithm. two-tailed t-test was applied to the manual and accelerometer data using the video data as the reference. filtering, threshold analysis standard error of the estimate for kick-count validity, expressed as a coefficient of variation (CV) typical error of the measurement for reliability for kick count and kick rate filtering, threshold analysis Kick-count validity was established between the inertial sensor and kicks counted manually from underwater video footage. coefficient of variation (CV) between the sensor and video footage
Filtering and signal processing algorithms (zero crossing) of acceleration data Difference between the real-time (acceleration data) and hand-computed stroke rates (video). Mean errors for each style. ICC: two trials for each participant to test the reliability of stroke rate estimation. Time integration, dynamic biomechanical constraint, geometric moving average change detection algorithm Principal Component Analysis Spearman’s correlation (rho = 0.94, P < 0.001) accuracy (
Journal of Sports Sciences Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/rjsp20
Wearable inertial sensors in swimming motion analysis: a systematic review a
b
c
Fabricio Anicio de Magalhaes , Giuseppe Vannozzi , Giorgio Gatta & Silvia Fantozzi
ad
a
Department of Electrical, Electronic, and Information Engineering “Guglielmo Marconi”, University of Bologna, Bologna, Italy b
Interuniversity Centre of Bioengineering of the Human Neuromuscoloskeletal System, Department of Motor, Human and Health Sciences, Universita’ di Roma Foro Italico, Rome, Italy c
Department for Life Quality Studies, University of Bologna, Bologna, Italy
d
Health Sciences and Technologies – Interdepartmental Center for Industrial Research, University of Bologna, Bologna, Italy Published online: 30 Oct 2014.
To cite this article: Fabricio Anicio de Magalhaes, Giuseppe Vannozzi, Giorgio Gatta & Silvia Fantozzi (2014): Wearable inertial sensors in swimming motion analysis: a systematic review, Journal of Sports Sciences, DOI: 10.1080/02640414.2014.962574 To link to this article: http://dx.doi.org/10.1080/02640414.2014.962574
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Journal of Sports Sciences, 2014 http://dx.doi.org/10.1080/02640414.2014.962574
REVIEW ARTICLE
Wearable inertial sensors in swimming motion analysis: a systematic review
FABRICIO ANICIO DE MAGALHAES1, GIUSEPPE VANNOZZI2, GIORGIO GATTA3 & SILVIA FANTOZZI1,4 Department of Electrical, Electronic, and Information Engineering “Guglielmo Marconi”, University of Bologna, Bologna, Italy, 2Interuniversity Centre of Bioengineering of the Human Neuromuscoloskeletal System, Department of Motor, Human and Health Sciences, Universita’ di Roma Foro Italico, Rome, Italy, 3Department for Life Quality Studies, University of Bologna, Bologna, Italy and 4Health Sciences and Technologies – Interdepartmental Center for Industrial Research, University of Bologna, Bologna, Italy
Downloaded by [University of Ottawa] at 19:07 23 November 2014
1
(Accepted 1 September 2014)
Abstract The use of contemporary technology is widely recognised as a key tool for enhancing competitive performance in swimming. Video analysis is traditionally used by coaches to acquire reliable biomechanical data about swimming performance; however, this approach requires a huge computational effort, thus introducing a delay in providing quantitative information. Inertial and magnetic sensors, including accelerometers, gyroscopes and magnetometers, have been recently introduced to assess the biomechanics of swimming performance. Research in this field has attracted a great deal of interest in the last decade due to the gradual improvement of the performance of sensors and the decreasing cost of miniaturised wearable devices. With the aim of describing the state of the art of current developments in this area, a systematic review of the existing methods was performed using the following databases: PubMed, ISI Web of Knowledge, IEEE Xplore, Google Scholar, Scopus and Science Direct. Twenty-seven articles published in indexed journals and conference proceedings, focusing on the biomechanical analysis of swimming by means of inertial sensors were reviewed. The articles were categorised according to sensor’s specification, anatomical sites where the sensors were attached, experimental design and applications for the analysis of swimming performance. Results indicate that inertial sensors are reliable tools for swimming biomechanical analyses. Keywords: swimming, biomechanics, inertial sensors
Introduction Currently, minute differences in the race times of finalists and winners of swimming events have forced athletes, coaches and sport scientists to work harder to get a “fraction of a second” improvement (James, Davey, & Rice, 2004). The use of contemporary technology plays an important role in supporting performance enhancement from a biomechanical point of view by allowing detailed analysis of motion to be carried out and thus refine swimming technique (Pansiot, Lo, & Guang-Zhong, 2010). The analysis of in-water movements can improve the understanding of swimming biomechanics, enabling swimmers to perform closer to their full potential
(Callaway, Cobb, & Jones, 2009). Systematic biomechanical analysis identifies technical errors and monitors performance progress (Daukantas, Marozas, & Lukosevicius, 2008) of the swimming technique, which provides individuals with a more efficient swimming technique. Traditionally, motion measurements during swimming have been performed using video analysis method as gold standard (Ceseracciu et al., 2011; McCabe, Psycharakis, & Sanders, 2011; McCabe & Sanders, 2012; Vezos et al., 2007). Therefore, the use of underwater video cameras for recording, modelling and refining swimming strokes has become a common approach for elite swimmers (Callaway et al., 2009). This method requires a
Correspondence: Giuseppe Vannozzi, IUC-BOHNES, Department of Motor, Human and Health Sciences, Universita’ di Roma Foro Italico, Rome, Italy. E-mail: [email protected] © 2014 Taylor & Francis
Downloaded by [University of Ottawa] at 19:07 23 November 2014
2
Fabricio Ancio de Magalhaes et al.
tracking procedure for estimating the position of given points of interest (typically anatomical landmarks) using algorithms specifically developed to extract performance-related biomechanical variables (Ceccon et al., 2013; Pansiot et al., 2010). Marker tracking is the most accurate and comprehensible method for swimming motion analysis, which provides relevant data to coaches and athletes (Callaway et al., 2009); however, making this process fully automatic is difficult. Furthermore, the tracking of markers is time-consuming, error-prone due to water turbulence, and requires intensive data processing (Gourgoulis et al., 2008; Wilson et al., 1999). Moreover, a number of synchronised cameras are required to achieve a complete 3D kinematic description, increasing the costs of the procedure. Furthermore, video analysis can only be performed off-line, which delays the process of providing real-time feedback by the coach to the swimmer to address technical errors. An automatic tracking procedure was recently developed (Magalhaes et al., 2013; Olstad, Zinner, Haakonsen, Cabri, & Kjendlie, 2012) to minimise the time-consuming processing phase. Even though reflective markers and dedicated cameras were used (Olstad et al., 2012), concerns still existed for video-based motion analysis. Following the current trend of quantitative human movement analysis in different application contexts, inertial measurement units (IMUs) worn by the athletes have been proposed as an alternative tool for in-field sports performance analysis to overcome the limitations of video-based methods. Swimming biomechanics has been strongly oriented toward the use of IMU technology due to the above-mentioned difficulties in measuring the movement of swimmers in the water by means of common video-based methods. Wearable IMU sensors do not require a complex measurement setup in the swimming pool and are considered swimmer-centric (Pansiot et al., 2010). The current advent of micro-electro-mechanical systems technology provides practical IMU applications by reducing both the dimension and power consumption of the devices. This technology allows various setups and positioning on the swimmer’s body. Research in this field has provided evidence of the application of IMU sensors for monitoring human movement in sport. Furthermore, these sensors have meaningful advantages allowing reliable measurements in different environments, such as in the water, and providing specific environmental advantages that can improve data acquisition. Several authors have examined the movement patterns and sports techniques using accelerometers, gyroscopes and, eventually, magnetometers (Caruso, 1997; Jones, Jones, & Nenadic, 2013). Following the
emerging trends of this technology, swimming science first incorporated the use of accelerometers and, more recently, the gyroscopes; therefore, this review focused only on inertial sensors. For example, accelerometry data feedback given to athletes provided kinematic and performance consistency improvements in rowing (Anderson, Harrison, & Lyons, 2005). A wide range of areas were interested, including: ambulatory measurements, physical activity assessment, gait analysis, and evaluation of athletic performance (Callaway et al., 2009; Davey, Anderson, & James, 2008). Therefore, the use of IMUs has been shown to be an overall effective tool for monitoring human movement patterns, eventually including additional wearable devices such as pressure, temperature and humidity sensors. An increasing range of inertial sensors and protocols have been proposed for swimming performance assessment, and it is of interest to examine how they were used and how well they performed with respect to traditional movement analysis techniques. The increasing body of literature on this topic has not been reviewed. Therefore, the purpose of this article was to provide a systematic review regarding the current status of inertial and magnetic sensors for swimming performance assessment. The main objective of this review was to provide a framework to fully exploit the recent advances in miniaturised wearable technologies to obtain biomechanical data related to sport performance. Methods Review questions A systematic review of the literature regarding the inertial sensor-based performance assessment was carried out to address the following questions: (1) What are the existing inertial sensors applications for swimming motion analysis? (2) What type of inertial sensor was used? (3) Where were the inertial sensors attached? (4) How were the experiments conducted? (5) How was the performance evaluated using this approach? Article selection The research method was based on a systematic search for publications in the following scientific databases: PubMed, ISI Web of Knowledge (Science Citation Index Expanded), IEEE Xplore, Google Scholar, Scopus and Science Direct. The search included all journal and conference papers published before August 2013. These electronic search engines/databases were chosen as the most common databases related to methodological issues in applied biomechanics, covering most of the related literature in engineering, medicine and
Downloaded by [University of Ottawa] at 19:07 23 November 2014
Wearable inertial sensors in swimming biomechanics in sports. The searched keyword string was “(swimming OR swim OR swimmer) AND (inertial sensors OR inertial sensor OR accelerometer OR accelerometers OR acceleration OR accelerometry OR gyroscope OR gyroscopes OR gyroscopic OR magnetic sensor OR magnetometer OR IMU OR sensor fusion OR MEMS OR inertial tracking devices OR body worn sensors OR pervasive computing OR WSN OR INS OR WIMU OR BSN)” appearing in the title, abstract and keyword fields of the articles. Common acronyms used in the context of inertial sensors technology were included in the search string such as MEMS (Micro ElectroMechanical Systems), WSN (Wireless Sensor Network), INS (Inertial Navigation System), WIMU (Wireless Inertial Measurement Unit), BSN (Body Sensor Networks). Only English language articles were considered, and the initial total number of articles identified from all databases was 65 (Medline 28, ISI 10, IEEE Xplore 16, Google Scholar 12, Scopus 6, Science Direct 13), including journal articles, conference proceedings and book chapters. As an initial search strategy, both the title and abstract were first identified and read carefully to establish a first selection of articles. Relevance of the article was evaluated based on the pertinence to the topic “measurement of swimming biomechanics”. When it was not clearly indicated whether an article should be included or not, the complete article was retrieved from the local university libraries and fully reviewed by the authors. Considering also the deletion of unrelated and duplicated articles, 27 full papers were ultimately included in this review. The inclusion criteria were as follows: (1) the study dealt with human swimming motion analysis, and (2) the study involved accelerometers, gyroscopes and/or magnetic sensors as main measuring devices. The following parameters were extracted by the review team: the participants’ characteristics, anatomical region involved in measurement, study procedures, type of sensor/portability, biomechanical variables measured, data processing methods and statistical analyses, gold standard measurement system, primary outcomes of accuracy and/or reliability. Following this grid, all the 27 research articles were organised as shown in Table I.
Results Applications of inertial and magnetic sensing in swimming Developing and testing specific protocols to perform a 3D swimmers’ kinematic analysis using IMUs has been the aim of several authors. The validation of sensors has been widely carried out in running (Bergamini et al., 2012; Lee, Mellifont, & Burkett,
3
2010) and jumping (Picerno, Camomilla, & Capranica, 2011). However, only a few swimming validation trials exist (Dadashi, Crettenand, Millet, & Aminian, 2012; Davey et al., 2008; Stamm, James, & Thiel 2013). Most of the studies are focused case studies, as reported in Table I. Based on the information contained in Table I, the sport biomechanist can select the appropriate sensor (accelerometer or gyroscope or both) and the relevant variables to estimate. The first attempts in using inertial sensors in swimming referred to the phase analysis, a fundamental prerequisite for each biomechanical analysis in sports. The first study using IMUs in swimming was carried out by Ohgi, Yasumura, Ichikawa, and Miyaji (2000). They investigated the relationships between the wrist acceleration data and hand movement under the Maglischo’s stroke pattern definition (Maglischo, 1993). Examining the tri-axial wrist acceleration of front crawl strokes, they verified the presence of a repeatable pattern for each stroke style that led to a feasible stroke phase identification (entry, down sweep, out sweep and in sweep). In addition, Ohgi, Ichikawa, Homma, and Miyaji (2003) using the same approach discriminated the different phases of the breaststroke style (recovery, in-sweep and out-sweep). Furthermore, IMUs output could be used to discriminate among the different swimming styles. Slawson et al. (2008), Siirtola, Laurinen, Roning, and Kinnunen (2011) and Hou (2012) reported that the four competition swim styles could be extracted from accelerometers worn on the wrist or the upper-back. Pansiot et al. (2010), by deriving pitch and roll angles, were able to detect the type of stroke and the wall push-offs by means of a single miniaturised accelerometer-based sensor mounted on the swimmer’s goggles. A similar method was used by Vannozzi et al. (2010) in the analysis of the four swimming strokes using gyroscope data obtained using a single waist-mounted IMU. The characteristics of the swimmer’s stroke motion can be acquired using IMUs easily and continuously (Ohgi & Ichikawa, 2002; Ohgi, Ichikawa, & Miyaji, 2002) during the entire length of the pool, instead of focusing on a single stroke as typically allowed by traditional video analysis. Although several authors made attempts to insert IMUs in the swimming environment, only a few quantitative parameters were extracted from the registered signals, and a few processing methods were used for that purpose. Typically, the swim data analyses have been carried out with a personal computer using off-line algorithms after the swimming exercise. This represents a very useful procedure for specific technical analyses to be performed after the competition or the physical training; however, the long processing time needed typically does not allow the coaches to give their immediate
Beanland, Main, Aisbett, Gastin, and Netto (2013) Chakravorti, Le Sage, Slawson, Conway, and West (2013) Dadashi et al. (2012)
Dadashi, Crettenand, et al. (2013) Dadashi, Arami, et al. (2013) Davey et al. (2008) Fulton, Pyne, and Burkett (2009b)
Fulton, Pyne, and Burkett (2009a)
Fulton, Pyne, and Burkett (2011)
Hagem, Thiel, O’Keefe, and Fickenscher (2013) Hou (2012)
James et al. (2011)
Le Sage et al. (2010b)
2
5
9
10
11
13
14
12
7 8
6
4
3
Bächlin and Tröster (2012)
Source
1
Serial nos
1
3
2
2
1
4
1 1
2
3
1
1
1
4
Units
3D acc 2D gyro
3D acc; temperature, humidity and pressure sensors 3D acc 3D gyro
3D acc
3D acc 1D gyro
3D acc 1D gyro
3D acc 3D gyro 3D acc 3D gyro 3D acc 3D acc 1D gyro
3D acc 3D gyro
3D acc
3D acc
3D acc
Type
Not described
Not described
52 × 33 × 11 mm3
Not described
18 g
Not described
50 Hz
100 Hz
Sealing
Participant
MiniTraquaTM, Version 5, Cooperative Research Centre for Micro-technology, Australian Institute of Sport Not described
Plastic casing
Cited: James et al. (2004) Plastic casing
Not described
Hermetical sealing the plastic bags
Hermetical sealing the plastic bags
In-house custom-built waterproof package 1 swimmer (double membrane), known as “sealedfor-life solution” The prototype node was packaged to ensure 1 swimmer it was waterproof for the application (no further detail provided). (continued )
11 swimmers (male and females)
1 swimmer
7 elite and 19 recreational swimmers 7 trained swimmers (5 M, 2 F) 7 trained swimmers (4 M, 3 F) 6 elite swimmers 14 Paralympic swimmers (8 M, 6 F) 12 Paralympic swimmers (8 M, 4 F) 12 Paralympic swimmers (9 M, 3 F)
Water-tightness: plastic foil. Wrist: inside a 18 swimmers transparent plastic tube fixed using velcro (7 competitive, 8 fasteners. recreational and 3 occasional) Not described 21 swimmers (high level at least 5 sessions per week Not described 2 swimmers
Range between 5 Not described and 50 Hz
MiniTraquaTM, Version 5, Cooperative 100 Hz Research Centre for Micro-technology, Australian Institute of Sport 27 × 19 mm Not described Not described
100 Hz
52 × 33 × 11 mm3
100 Hz
500 Hz
500 Hz
Not described
100 Hz
256 Hz
Sample frequency
Not described 100 Hz
20.7 g
Not described
Not described
34 g
Weight
Cited: James et al. (2004) Cited: Fulton et al. (2009a)
Physilog®, BioAGM, CH
Physilog®, BioAGM, CH
Physilog®, BioAGM, CH
Not described
Not described
36 × 42 × 12 mm3
Sizes
Type and specification of sensors
Table I. List of the selected 27 articles organised with the indication of the first author and containing the different details related to sensor type and specification, number and type of participants, sensor location body area, analysed stroke type, spatiotemporal and biomechanical variables investigated, gold standard reference measurement, eventual data analysis methods and adopted statistics. Acronyms for the different styles are the following: front crawl (FC), breaststroke (BrS), backstroke (BaS) and butterfly (BF).
Downloaded by [University of Ottawa] at 19:07 23 November 2014
4 Fabricio Ancio de Magalhaes et al.
Le Sage et al. (2011b)
Le Sage et al. (2012)
Lee et al. (2010)
Lee, Ohgi, and James (2012)
Nakashima, Ohgi, Akiyama, and Kazami (2010) Ohgi et al. (2002)
Ohgi et al. (2003)
Pansiot et al. (2010) Siirtola et al. (2011)
Slawson et al. (2008)
Slawson, Justham, and Conway (2012) Stamm, Thiel, Burkett, and James (2011)
Stamm et al. (2013)
15
16
17
18
19
20
21
22 23
24
25
27
26
Source
Serial nos
Table I. (Continued).
1
1
1
1
1 2
1
1
1
1
1
1
1
Units
3D acc 3D gyro
3D acc 2D gyro 3D acc
3D acc 3D acc; temperature, humidity and pressure sensors 3D acc
2D acc
3D acc
3D acc 3D gyro
3D acc
3D acc 3D gyro
3D acc 2D gyro 3D acc
Type
190 Hz
100 Hz
200 Hz
Not described
25 Hz
Sample frequency
Not described
20 g
53 × 33 × 10 mm3
Not described
Not described
8g 18 g
62 g
100 Hz
100 Hz
Not described
50 Hz
50 Hz 5, 10, 25, 50 Hz
Not described
Prototype I: 50 g 128 Hz Prototype II: 78 g
Not described
Cited: Davey et al. (2008)
Not described
Not described
110 g
Weight
Not described
90 mm × 40 mm
Not described
Not described Not described
Not described, but smaller than 44 × 37 × 20 mm3
Prototype I: 88 × 21 mm Prototype II: 141.8 × 23.2 mm
Not described
Cited: Davey et al. (2008)
Not described
Not described
15 × 9 cm
Sizes
Type and specification of sensors
Sealing
1 swimmer
Participant
1 amateur level triathlete 1 male elite triathlete
(continued )
Not described but it can be seen in the figure 1 recreational 1 that the sensor was put in a plastic (or swimmer acrylic) housing. Waterproof casing 17 swimmers (8 junior elite and 9 retired elite)
Not described
Not described
2 top-level college swimmers (experiment 1) and 5 (4 male and 1 female) triathlon athletes (experiment 2) Not described but it can be seen in the figure 2 swimmers 1 that the sensor was put inside of a wristwatch. Rubber latex skin 1 swimmer Not described 11 amateurs and professionals
capsulated by aluminium alloy cylinder (200 m water resistant)
12 swimmers (8 males and 4 females) 6 swimmers (3 male and 3 female) 2 elite athletes (a swimmer and a triathlete) The wristwatch-style sensing unit is cased in 1 swimmer a waterproofed housing.
The sensor, battery and an antenna were packaged into a waterproof AquaPac Not described but it can be seen in the figure 4 that the sensor was packed in a plastic (or rubber) housing. Not described but it can be seen in the figure 1 that the sensor was housed in a plastic (or acrylic) cubic recipient. Cited: Davey et al. (2008)
Downloaded by [University of Ottawa] at 19:07 23 November 2014
Wearable inertial sensors in swimming 5
Upper back lower back right wrist
head
Lower back
Lower back (sacrum)
both forearm Lower back (sacrum)
Right arm, right leg
Lower back (sacrum)
dominant kicking leg
Both thighs both shanks
– – –
–
–
–
– –
–
–
–
– –
2
3
4
5
6
7
8
9
Body area
1
Serial nos
Table I. (Continued).
Only flutter kick
Only flutter kick
All
BrS
FC
FC
All
FC, BrS, BF
FC
Stroke type
– –
– – –
– – –
–
– –
–
– – –
– – – – – – –
–
– –
–
kick patterns kick-detection
kick count kick rate kick patterns
push-off stroke type stroke count metrics
–
–
–
Stroke phases (arm and – leg)
stroke phases inter-arm coordination
instantaneous velocity
stroke counts stroke rates lap counts
start (wall-push-off) – strokes end (wall-strike) Average velocity time per stroke (TPS) distance per stroke (DPS) stroke counts –
Variables
– –
–
–
– –
Video data (underwater)
Video data (underwater)
–
– –
–
– –
Hand timed and video data – (underwater) –
tethered apparatus – (SpeedRT®, ApLab, Rome, Italy). – – – Video data (2 underwater – cameras, 50 Hz; 189 – videos) Video data (2 underwater – cameras, 50 Hz; 168 – videos)
Video camera (2 fixed and 1 trolley mounted cameras; 25 Hz) Video data (1 underwater camera and 1 high speed video camera)
Normal stopwatch and real – accelerometer data –
Gold standard
Data analysis methods and statistics
(continued )
Phase detection algorithm (Hidden Markov Model) Mean absolute deviation, mean and standard deviation of the error; repeated measure ANOVA (difference not influenced by either of athlete or trial factors), sensitivity (93.5% arm phase, 94.4% leg phase) and specificity (96.2% arm phase, 97.2% leg phase) Maximum and minimum detection, threshold analysis, zero crossing algorithm. two-tailed t-test was applied to the manual and accelerometer data using the video data as the reference. filtering, threshold analysis standard error of the estimate for kick-count validity, expressed as a coefficient of variation (CV) typical error of the measurement for reliability for kick count and kick rate filtering, threshold analysis Kick-count validity was established between the inertial sensor and kicks counted manually from underwater video footage. coefficient of variation (CV) between the sensor and video footage
Filtering and signal processing algorithms (zero crossing) of acceleration data Difference between the real-time (acceleration data) and hand-computed stroke rates (video). Mean errors for each style. ICC: two trials for each participant to test the reliability of stroke rate estimation. Time integration, dynamic biomechanical constraint, geometric moving average change detection algorithm Principal Component Analysis Spearman’s correlation (rho = 0.94, P < 0.001) accuracy (
