Content uploaded by Stephen Barrett
Author content
All content in this area was uploaded by Stephen Barrett on Nov 18, 2020
Content may be subject to copyright.
1
Ahead of print DOI: 10.2478/hukin-2021-0010
A new foot-mounted inertial measurement system in soccer:
reliability and comparison to global positioning systems for
velocity measurements during team sport actions
Mark Waldron1,2*, Jamie Harding3, Steve Barrett4, Adrian Gray2
1Swansea University, College of Engineering, A-STEM, Swansea, UK.
2University of New England, School of Science and Technology, NSW, Australia.
3St Mary’s University, Faculty of Sport, Health and Applied Sciences, London, UK.
4Hull City Tigers FC, East Riding of Yorkshire, UK.
* = corresponding author
Dr Mark Waldron
College of Engineering
Engineering East (A120)
Swansea University,
Swansea,
UK.
Email: mwaldro4@hotmail.com
Phone: +44 7774004973
Conflict of interest: The authors have no conflict of interest to declare
Acknowledgments: The results of the current study were presented at the United Kingdom Strength and
Conditioning Association annual conference 2019, Milton Keynes. No financial support was received for this
study.
Abstract
The aims of this study were to i) compare a foot-mounted inertial system (PlayerMaker™) to three
commercially available Global Positioning Systems (GPS) for measurement of velocity-based metrics during
team sport movements and ii) evaluate the inter-unit reliability of the PlayerMaker™. Twelve soccer players
completed a soccer simulation, whilst wearing a PlayerMaker™ and three GPS (GPS#1, #2 and #3). A sub-
sample (n = 7) also wore two PlayerMaker™ systems concurrently. The PlayerMaker™ measured higher (p <
0.05) total distance (518 ± 15 m) compared to GPS#1 (488 ± 15 m), GPS#2 (486 ± 15 m), and GPS#3 (501 ± 14 m).
This was explained by greater (p < 0.05) distances in the 1.5-3.5 m/s zone (356 ± 24 m vs. 326 ± 26 m vs. 324 ±
18 m vs. 335 ± 24 m) and the 3.51-5.5 m/s zone (64 ± 18 m vs. 35 ± 5 vs. 43 ± 8 m vs. 41 ± 8 m) between the
PlayerMaker™, GPS#1, GPS#2 and GPS#3, respectively. The PlayerMaker™ recorded higher (p < 0.05)
distances while changing speed. There were no systematic differences (p > 0.05) between the two
PlayerMaker™ systems. The PlayerMaker™ is reliable and records higher velocity and distances compared to
GPS.
Key words: soccer, motion analysis, player tracking.
2
Introduction
Wearable sensors are routinely used to track the movements of athletes during training and
competition (Cummins et al., 2013; Mallo et al., 2015). Team sports use micro-technology, containing both
micro-electromechanical systems (MEMSs) and global positioning systems (GPSs), worn in a vest between the
player’s scapulae to track athletes’ gross movements (Barrett et al., 2016). Using GPSs, time motion analysis
data, such as total distance covered, distance in selected velocity zones, accelerations and decelerations are
often reported to describe the external load of training or competition (Cummins et al., 2013). While most
micro-technology devices used in team sports contain MEMSs, output from these sensors provides segregated
performance variables and does not contribute to the calculation of velocity-based metrics (Malone et al., 2017).
The validity and reliability of a number of commercially available systems in measuring velocity-
based metrics have been described (Akenhead et al., 2014; Buchheit et al., 2014; Coutts and Duffield, 2010;
Varley et al., 2012), thus providing an understanding of the application and limitations of GPSs in a team
sports context. A common limitation among these studies has been the validity and reliability of GPS devices
to measure high- and variable-velocity movements in smaller areas. Notably, the accuracy of rapid
accelerations (> 3 m/s2) is consistently compromised when using GPS devices of varying specifications
(Akenhead et al., 2014; Buchheit et al., 2014). Low sampling rates (Varley et al., 2012), the positioning of the
device (Barrett et al., 2014, 2016), quality of the satellite signal (Karaim and Aboelmagd, 2018) and
inconsistencies within the data processing (Buchheit et al., 2014; Varley et al., 2017), have been suggested to
contribute to the error of GPS devices.
GPS devices typically underestimate criterion measures of mean velocity or distance (Duffield et al.,
2010; Vickery et al., 2014), but possess generally acceptable internal consistency, such that typical changes in
performance can be identified. However, coefficients of variation (CV) ranging between 20-78% have been
reported for the inter-unit reliability (i.e. agreement between two devices) of the same manufacturer and
model (Coutts and Duffield, 2010; Thornton et al., 2019). As a result, individual assignment of devices among
a squad of players has been recommended, as well as limiting between-player comparisons.
Alternative measurement methods are capable of tracking identical velocity-based metrics in a team
sports context, yet adopt a different technological approach. Inertial measurement units (IMUs) provide one
example and can be fitted about the person to monitor performance (van der Kruk and Reijne, 2018). Wearable
IMUs comprise accelerometers, gyroscopes and can include a magnetometer. Measurements of raw
acceleration and angular velocity are recorded during movement to detect temporal gait events on a stride-
by-stride basis (Yang et al., 2011). Integration of these data can determine velocity and orientation of various
body parts, respectively, depending on the selected anatomical placement of the IMU, without the
requirement of the GPS. IMUs are commonly fitted to the lower-limbs (shank or foot) for the purposes of gait
analysis, with early attempts to validate linear horizontal velocity reporting errors between ~ 2 and ~ 5%
(Hausswirth et al., 2009; Yang et al., 2011). However, most IMUs have not been specifically designed to
quantify movements observed among team sports players during training or competition. Given the
acceptable criterion validity of existing IMUs (Roell et al., 2018), this limitation could be overcome by a foot-
mounted unit, assuming the underlying algorithms are suitably designed to accommodate the complexity of
team sports movements and potential drift errors (Takeda et al., 2014).
It is somewhat surprising that integrated IMUs are not more frequently used for measuring velocity-
based metrics in team sports, since they can be used in- or outdoors and have fewer potential sources of
measurement error (van der Kruk and Reijne, 2018). However, the positioning of GPS-micro-technology
devices between the scapulae limits the scope of IMU output, whereas foot-mounted IMUs are capable of
recording the inertial motion and 3D orientation of individual limbs or limb segments with high accuracy
(O'Reilly et al., 2018). For example, Zaferiou et al. (2017) identified individual gait characteristics, such as
horizontal ground reaction, foot trajectories and footfall duration, alongside horizontal velocity, of high- and
low-level performers during running agility tasks using IMUs fixed to the dorsal aspects of both feet. While
such detailed analyses would be useful in a team sports context, the capacity of foot-worn IMUs to reliably
measure the velocity of players during team sports movements is unknown. Furthermore, to date, there has
been no direct comparison of foot-mounted, sports-specific IMUs to the most commonly adopted form of
velocity measurement (GPS-micro-technology), while performing movement patterns that simulate the
demands of a particular team sport. Thus, the primary aim of this study was to compare two methods of
velocity measurement; a foot-mounted IMU that has been specifically developed for soccer performance
3
(PlayerMaker™) to three commercially available GPS devices for the measurement of horizontal velocity
during team sport movements. The second aim was to evaluate the inter-unit reliability of the PlayerMaker™
system.
Methods
Design
Participants visited the testing facility twice. After initial familiarisation (visit 1) to the protocol and
the instrumentation, participants came to the research facility on a separate day to perform 5-min and 20-s of
a soccer-specific, multi-directional intermittent movement protocol (SAFT90; Barrett et al., 2013). Participants
were fitted with a foot-worn IMU (PlayerMaker™) and three separate GPS devices, which are most commonly
used in elite team sports, but were anonymised for the current article. GPS#1 sampled at 18 Hz, GPS#2 and
GPS#3 sampled at 10 Hz. Fitting of the IMU (PlayerMaker™) and the three GPS devices facilitated a method
comparison of velocity-based performance metrics.
Participants
Twelve elite League 1 academy-level soccer players (age 15 ± 3, range 11-18 years; body mass 54.5 ±
14.9 kg) provided written informed consent and, where necessary, parental assent to participate in this study.
Institutional ethical approval was provided for this study (SMEC_2018-19_015). Participants were informed
of the benefits and risks of the investigation prior to signing the institutionally approved informed consent
document to participate in the study
Procedures
The SAFT90 was designed to replicate the movement demands of English Championship-level soccer,
based on time-motion analysis of match play, dictating the actions and pace of the movement via pre-recorded
audio instructions. The SAFT90 course is 20 m in length, comprising varying intensity (stand, walk, jog, stride
and sprint) and multi-directional actions (forward, backward and sideward locomotion) (Figure 1). The
participants’ intended route through the cone-marked course is described in Figure 1, totalling a distance of
484 m. However, given the deviation from the intended route of all participants while completing the course,
this was not used to determine criterion distance. Participants were instructed to follow the audio cues in the
same way as their familiarisation trials and ensure maximal effort during the sprint sections. All trials were
performed on a flat, well-groomed, real grass pitch, under fair weather conditions, with participants wearing
studded soccer boots and a standardized training kit issued by the soccer team. Testing was performed on the
middle of the pitch to minimise the effects of the local built environment on GPS signal quality.
Instrumentation and data collection
The same three GPS devices where tightly fitted to each participant using custom-designed neoprene
vests, placing the units 2.5 cm apart, in parallel between the scapulae (Thornton et al., 2019). The GPS devices
were kept in the same position on each participant for all trials (Left unit - GPS#2; Central unit - GPS#3; Right
Unit - GPS#1). All GPS devices were activated 20-min prior to testing to ensure that the maximal number of
satellite connections were established. There was an acceptable number of identified satellites and horizontal
dilution of precision for GPS#3 (10 ± 2 and 0.8 ± 0.1, respectively), GPS#2 (16 ± 1 and 0.7 ± 0.1, respectively) and
GPS#1 (13 ± 2 and 0.4 ± 0.1, respectively). Participants were instructed to stand motionless prior to each trial
to assist with data synchronisation. All devices were synchronized post-hoc using local time, as measured on
board the GPS devices. While this was typically sufficient, the raw velocity traces were visually checked to
verify temporal similarity, using the continuous acceleration from quiet standing as an indication of the
initiating movement. All data were uploaded to each manufacturer’s software package and ‘raw’ (unfiltered
10 Hz) velocity traces were extracted into Microsoft Excel 2016 (Microsoft Inc. WA, US.). These data were
subsequently interpolated and filtered (see data analysis).
Participants were simultaneously fitted with a PlayerMaker™ system, which is a footwear sensor unit,
housed within custom silicone straps, securely fixed over each football boot, sitting on the lateral aspect of the
calcanei (Figure 2). The PlayerMaker™ was activated 10 min prior to the session and required no calibration
by the user prior to data collection. Participants were instructed to stand motionless prior to each trial to assist
with data synchronisation. The PlayerMaker™ sensor is an IMU, comprising a 3-axis 16 g accelerometer and
3-axis gyroscope (MPU-9150, InvenSense, California, USA) for measurement of accelerations and angular
velocity of each foot during gait, respectively.
4
The PlayerMaker™ system calculates whole-body velocity-based metrics using data generated by the
on-board MEMS, which utilizes a combination of proprietary gait tracking and foot-based event detection
algorithms. In brief, the soccer-specific gait tracking algorithm permits detection of the orientation and
translation of the participants‘ limbs during gait cycles, while the event detection algorithm identifies key
events during gait (heel strike, toe-off, zero-velocity, zero height, non-gait pattern). The microprocessor
receives accelerometer and gyroscope data, from which orientation, velocity and position vectors are
determined utilizing a Kalman Filter together with gait events update (Figure 3). The resulting output can
provide a number of metrics germane to soccer performance; however, velocity profiles were extracted for
comparison to the GPS devices.
Inter-unit reliability
In a sub-analysis, seven of the participants (age 15 ± 2 years; body mass 57.5 ± 12.5 kg) wore two
PlayerMaker™ systems concurrently (two on each foot) during the completion of the SAFT90 to facilitate inter-
unit reliability analysis. The units were fitted adjacent to one another on the lateral aspect of the calcaneus.
The same two units were worn throughout the study.
Data analysis
Raw velocity files were downloaded and exported to Microsoft Excel 2016 (Microsoft Inc. WA, US) for
further analysis. All velocity traces were interpolated to 25 Hz signals to facilitate temporal comparisons and
to remove sampling frequency differences between the devices. 25 Hz represents the down-graded sampling
rate of the PlayerMaker™ from 1000 Hz and was deemed a suitable frequency for comparative purposes,
owing to the typical centre of mass (COM) movement frequency during human locomotion ~ 10 Hz (Welk,
2002) and the minimum sampling rate determined by the Nyquist principle. Given the reported differences in
output between software-derived and raw GPS velocity traces (Thornton et al., 2019), GPS raw data from
GPS#2 and GPS#1 devices were identically filtered using a zero-lag exponential filter (Malone et al., 2017;
Varley et al., 2017). The GPS#3 uses a median filter (personal communication) for velocity data, which we chose
to maintain as this represents the output of the device in practice. The PlayerMaker™ data were down sampled
from its origin of 1000 Hz to 25 Hz using a zero-lag Butterworth filter.
A total of 19 different time motion analysis variables were selected for analysis during the SAFT90,
based on those typically reported to describe soccer performance (Vigh-Larsen et al., 2018; Waldron and
Murphy, 2013): total distance (m); mean velocity (m/s); peak velocity (m/s); distance < 1.5 m/s (m); distance
1.5-3.5 m/s (m); distance 3.51-5.5 m/s (m); distance > 5.5 m/s (m); peak acceleration (m/s2); peak deceleration
(m/s2); acceleration count < 1.5 m/s2; acceleration count 1.5-3.5 m/s2; acceleration count > 3.5 m/s2; deceleration
count < -1.5 m/s2; deceleration count -1.5 - -3.5 m/s2; acceleration count < -3.5 m/s2; mean acceleration distance
(m); total acceleration distance (m); mean deceleration distance (m); total deceleration distance (m). The period
selected for acceleration or deceleration measurements was 0.3 s (Malone et al., 2017; Varley et al., 2017), with
minor accelerations (< 0.02 m/s2) removed to reduce low-intensity acceleration or deceleration counts. Peak
velocity was averaged over the fastest 0.3-s period during the SAFT90.
Statistical analysis
Systematic biases between the PlayerMaker™ system and each of the three GPS devices were
performed for all time motion analysis variables using paired t-tests (Atkinson and Nevill, 1998), with
statistical significance accepted at p < 0.05, with post-hoc Bonferroni adjustments. No statistical comparisons of
GPS devices were necessary to conduct. The 95% Limits of Agreement (95% LoA) were used to identify the
degree of systematic bias and random error (SD of the between system differences × 1.96) (Atkinson and
Nevill, 1998) between the PlayerMaker™ system and each of the three GPS devices. Reference to ‘total error’
hereafter refers to the sum of systematic and random error. The recommended checks for heteroscedasticity
(Atkinson and Nevill, 1998) were performed on the data generated during the SAFT90, with no significant
relationships found (p > 0.05). All statistical comparisons were performed in IBM SPSS (Software V24.0, IBM,
NY, USA), with statistical significance accepted at p < 0.05.
Results
PlayerMaker™ to GPS comparison during the Saft90 protocol
As presented in Tables 1 & 2 and Figure 4, there were differences (p < 0.05) between most
PlayerMaker™ and GPS variables during the SAFT90. However, there were no differences (p > 0.05) for peak
velocity comparisons, distance > 5.5 m/s between the GPS#1, GPS#3 and PlayerMaker. The PlayerMaker™
5
consistently recorded greater (p < 0.05) total distances compared to all GPS devices (mean range = 17 - 32 m
underestimation), as well as greater distance covered between 1.5-3.5 m/s (mean range = 20 – 32 m
underestimation) and 3.51-5.5 m/s (mean range = 22 – 30 m underestimation). The distance covered between
3.51-5.5 m/s produced the widest LoA with the GPS#1 device (total error of ~ 60 m). For total distance, the total
error between the GPS devices and the PlayerMaker™ ranged from ~ 25 to 50 m.
Similarly, peak accelerations (mean range = 0.3 – 1.75 m/s2 underestimation) and decelerations (mean
range = 0.6 – 1.56 m/s2 underestimation) were of smaller magnitude when measured by GPS devices compared
to PlayerMaker™. This translated to a total error of up to ~ 3.1 m/s2 between the PlayerMaker™ and the GPS
devices (Tables 1 & 2). However, the differences (mean systematic biases) were not all in the same direction,
with the GPS devices all registering a higher number of lowest intensity (< 1.5 m/s2) accelerations (mean range
= 29-130 overestimation) and decelerations (mean range = 111 and 195 overestimation for GPS#3 and GPS#1,
respectively). This trend was reversed for higher intensity accelerations. Total acceleration distance was higher
for all of the GPS devices (p < 0.05; range = 8 – 12 m), but mean acceleration distance was higher for the
PlayerMaker™ (p < 0.05; range = 0.02 – 0.08 m). Total deceleration (p < 0.05; range = 9 – 20 m) and mean
deceleration (p < 0.05; range = 0.02 – 0.17 m) distance was lower when measured by all GPS devices. Figure 5
shows the velocity profile of a representative participant during a 60-s segment of the SAFT90 to highlight the
above differences.
PlayerMaker™ inter-unit reliability during the Saft90 protocol
As presented in Tables 1 & 2, the PlayerMaker™ system was not systematically different to its
comparative unit for any variable (p > 0.05) during the SAFT90. The 95% LoA revealed that the low-intensity (<
1.5 m/s2) acceleration and deceleration counts were least reliable, with random variation of ± 21 and ± 16. The
random error of total distance between the two PlayerMaker™ systems was ± 1.28 m during the total 5-min,
20-s SAFT90 protocol. Figure 6 shows the velocity traces of two PlayerMaker™ representative systems during
a 75-s segment of the SAFT90.
Discussion
The first aim of this study was to compare the PlayerMaker™ system to three commercially available
GPS devices for the measurement of velocity-based metrics during team sports movements. We found a
number of differences between methods, with measurements such as total distance covered and distance in
three velocity zones higher in the PlayerMaker™ system. These differences were accompanied by larger peak
accelerations and peak decelerations, as well as a higher number of high-intensity velocity changes, measured
by the PlayerMaker™. The GPS devices tended to measure a larger number of low-intensity accelerations and
decelerations (particularly the GPS#1 and GPS#3 devices), leading to larger total acceleration distances, yet
lower mean acceleration and deceleration distances. Despite mean velocity discrepancies during the SAFT90,
there were no differences in peak velocity between all devices. It would appear that the change in method
(technological and anatomical differences) leads to differences in the quantification of time-motion analysis
data during intermittent team sports activity, with the IMU generally recording higher distances, velocities
and velocity changes compared to the GPS devices. These differences were not apparent when the
PlayerMaker™ system’s inter-unit reliability was assessed, demonstrating the consistency of these devices for
measuring team sports movement patterns.
GPS devices have a number of well-documented error sources, including orbital error, satellite clock
error, ionospheric error, tropospheric error and multipath and receiver noise (Karaim and Aboelmagd, 2018).
The accumulation of these errors, mixed with limited raw data sampling frequency, has led to underestimation
of criterion velocity and distance covered by ~ 10-30% (Duffield et al., 2010; Vickery et al., 2014). Whist we did
not have a reference system in the current study, it is worth recognising that the GPS devices also recorded
lower values for a number of velocity-based metrics compared to the PlayerMaker™ system. Spatio-temporal
gait characteristics measured by more rudimentary IMUs typically compare closely to gold-standard 3D
camera systems (Kluge et al., 2017), providing further confidence that the MEMS included in the
PlayerMaker™ are less susceptible to technical error. However, the complexity of velocity measurement
during team sports movements provides additional challenges to the calculation of whole-body velocity when
using IMUs. Indeed, while IMUs do not have any of the same limitations as GPS devices, a problem often
experienced is so-called ‘drift error’. Here, IMUs fitted to lower-limbs can measure velocity of body segments
by integration of raw acceleration; however, this leads to cumulative signal errors across time (i.e. drift). The
6
PlayerMaker™ system corrects for these errors using a Kalman Filter, whereby zero-velocity updates are used
to calibrate the sensors. While these approaches are well-known for simple locomotive patterns, such as linear
walking or running (Takeda et al., 2014), they are insufficient for estimating velocity when gait patterns are
more complex. The PlayerMaker™ uses advanced gait phase detection for soccer-specific movement, based
on a custom-built machine learning algorithm. The advanced gait phase detection process detects the zero-
velocity phases more efficiently and, in addition, detects zero-height phases, which usually occur during the
stance phase of gait. The aforementioned detections are used as input to the Kalman Filter and affect the
position and velocity estimation, thus controlling drift errors. These processes reduce the error of the system
and assist with the accuracy of the PlayerMaker™.
There are obvious differences in the technology and algorithmic approaches used to calculate velocity
between GPS and IMUs, which provide some explanation for the differences observed in the current study.
However, it is noteworthy that the anatomical placement of the devices is markedly different and will
influence the data generated during team sports movement patterns (Barrett et al., 2014, 2016). Measurement
of the COM (or spinal alignment thereof) does not reflect the entire movement of the lower limbs, leading to
discrepancies in movement detection between lower limbs and the COM during soccer specific activities
(Barrett et al., 2016; Nedelec et al., 2014). The larger difference of the PlayerMaker™ to the 484 m ‘intended
route’ perhaps demonstrates this, where the players’ foot placement will inevitably deviate from the body’s
centre line and cover greater distances.
The ability to monitor team sport activities using the GPS or IMUs has been historically difficult, due
to the random, intermittent and multi-directional nature of movements. Short, rapid and forceful movements
are necessary to efficiently increase or decrease velocity while completing complex movement patterns,
typically resulting in greater horizontal displacement of the ground contact points (i.e. the feet) relative to the
COM (Morrison et al., 2015). Based on this reasoning, the GPS unit placed between the scapulae and the foot-
mounted sensors will follow separate paths during team sports running patterns. Furthermore, the
PlayerMaker™ system is placed on two anatomical landmarks and must approximate horizontal velocity
across both limbs. Given the complex nature of movements required to perform the SAFT90, it is assumed that
the PlayerMaker™ measures a number of minor foot displacements, which would not register as clearly via a
single unit placed between the scapulae. This is, perhaps, more likely among well-trained athletes, who
typically attempt to maintain a narrower base of support and reduce turning arcs during change of direction
tasks, resulting in less displacement of the upper trunk, in favour of higher frequency ground contacts
(Zaferiou et al., 2017). This anatomical placement is important to the results of the current study and, alongside
the higher sensitivity of the PlayerMaker™, is likely to explain much of the variation between systems.
Based on the above reasons, the PlayerMaker™ system quantifies rapid changes in velocity during
soccer-based movements that will be unregistered by GPS devices, owing to a change in the method of
measurement (i.e. technology and anatomical placement). The change in method leads to greater distances
being reported, particularly in higher velocity zones, alongside larger mean acceleration or deceleration
distances. The higher acceleration distance measured by all GPS units is explained by higher frequency of very
low-intensity accelerations and might reflect the inability of the GPS signal to register higher-intensity velocity
changes compared to IMUs. This is more likely because, unlike the GPS-velocity, the PlayerMaker™ system
is not limited by a sampling rate, with a raw sampling rate of 1000 Hz. Thus, fast directional change with high
foot cadence can be readily tracked. These findings are of critical importance to soccer practitioners, as high-
intensity actions are associated with elite soccer performance levels (Waldron and Murphy, 2013) and
differentiate between positions (Vigh-Larsen et al., 2018). Moreover, the PlayerMaker™ registered greater
deceleration distances and high-intensity counts, which is consistent with the poorer validity of 10 Hz GPS
devices in measuring decelerations (11.3% CV), which are typically of larger magnitude compared to
accelerations (Varley et al., 2012). This is important, since decelerations are associated with increased muscle
soreness and impaired neuromuscular function following soccer matches (Nedelec et al., 2014) and, therefore,
require accurate quantification.
Among all of these measures, there were no systematic inter-unit differences for the PlayerMaker™
and the largest 95% LoA for distance covered being ~ 4.5 m (Table 2) across all velocity zones. Small total errors
in distance travelled would permit detection of a number of differences, including differences in match-
running performance of elite and sub-elite soccer players (Waldron and Murphy, 2013). There was, however,
a larger random variation for total acceleration distance (± 12.52 m; Table 2) between units, indicating slightly
7
larger noise for this variable that might preclude detection of more refined signal changes. However, this
random noise would still enable detection of changes in acceleration distance that might be important in
practice. For example, acceleration distance changes by a total of ~ 50 m across intensity zones between the
first and the second 15-min period of professional matches (Akenhead et al., 2014), which is much larger than
the inherent noise. Similarly, despite the systematic differences between methods and total error of up to ~ 25
m (GPS#1, Table 1), this noise is still within a ~ 50 m signal change and would, therefore, not significantly alter
the interpretation between methods. The disagreement between methods and random error within the
PlayerMaker™ would not become more noticeable until smaller changes in acceleration distance, such as the
5-10 m declines between final match periods, are necessary to quantify (Akenhead et al., 2014). Of interest, the
random error of deceleration distances was lower (~ 6 m; Table 2) for the PlayerMaker™, which we speculate
might be related to the motion of the ankle during these movements compared to accelerations. This is beyond
the scope of the current study, but is worthy of further investigation.
Soccer researchers and practitioners should, therefore, consider the current results in relation to their
desired outcomes. That is, the PlayerMaker™ system is more consistent between units compared to previous
GPS reports (Coutts and Duffield, 2010; Thornton et al., 2019) and will bias its measurement of whole-body
velocity towards the movement of the lower-limbs. This is arguably of greater importance to soccer
practitioners, since understanding of work done by the lower limbs during team sports movement patterns
has been incorporated into recent mechanical energetic models (Gray et al., 2018) and it is lower-limb measures
of muscle function that are often prioritised to determine exercise-induced fatigue in practice (McCall et al.,
2015). Furthermore, foot mounted IMUs have the capacity to measure an array of mechanical loading/gait
parameters, which have not been explored in the current study, but could be used by practitioners to assess
performance, asymmetries and neuromuscular function of players during training or competition.
Conclusions
Soccer practitioners should be aware of the differences between these two distinctly different methods
(GPS or foot-worn IMU) for the measurement of over-ground velocity. Distances covered in higher velocity
zones, peak accelerations and decelerations and high-intensity velocity changes are higher when measured
using foot-mounted IMUs (PlayerMaker™) compared to three commercially available GPS devices. Mean
velocity, but not peak velocity, also differs between these two types of technology. Practitioners can, therefore,
use foot-worn IMUs or GPS devices for tracking players during training and competition (depending on
governing body regulations), but understand that movement of the lower-limbs during short and rapid
changes will be directly incorporated into the PlayerMaker™ velocity measurement, while GPS velocity will
be based on displacements of a single sensor placed superior to the COM. If it is desirable for practitioners to
bias this measurement toward lower-limb movements, then the foot-worn system would be preferable. Those
working with soccer players should also consider that the PlayerMaker™ has inter-unit reliability that would
enable interchangeable interpretations for almost all time motion data. This would be useful for comparisons
between players, without concern over noise emanating from technical errors. Further research on the
PlayerMaker™ system to soccer performance will help understand its potential applications more thoroughly.
References
Akenhead R, French D, Thompson KG, Hayes PR. The acceleration dependent validity and reliability of 10 Hz
GPS. J Sci Med in Sport, 2014; 17: 562–566
Atkinson G, Nevill AM. Statistical methods for addressing measurement error (reliability) in variables relevant
to sports medicine. Sport Med, 1998; 26: 217-238
Barrett S, Guard A, Lovell R. SAFT90 simulates the internal and external loads of competitive soccer match-
play. In Science and Football VII: Proceedings of the Seventh World Congress on Science and Football,
Chapter: 15, Publisher: Routledge, Editors: H. Nunome, B. Drust and B. Dawson, pp.95-100, 2013
Barrett S, Midgley A, Lovell R. PlayerLoad™: reliability, convergent validity and influence of unit position
during treadmill running. Int J Sport Physiol Perform, 2014; 9: 945-952
Barrett S, Midgley AW, Towlson C, Garrett A, Portas M, Lovell R. Within-match Playerload™ patterns during
a simulated soccer match: potential implications for unit positioning and fatigue management. Int J
Sport Physiol Perform, 2016; 11: 135-140
8
Buchheit M, Al Haddad H, Simpson BM, Palazzi D, Bourdon PC, Di Salvo V, Mendez-Villanueva A.
Monitoring accelerations with GPS in football: time to slow down? Int J Sports Physiol Perform, 2014; 9:
442–445
Coutts A, Duffield R. Validity and reliability of GPS devices for measuring movement requirements in team
sports. J Sci Med Sport, 2010; 13:133-135
Cummins C, Orr R, O’Connor H, West C. Global positioning systems (GPS) and microtechnology sensors in
team sports: a systematic review. Sport Med, 2013; 43: 1025-1042
Duffield R, Reid M, Baker J, Spratford W. Accuracy and reliability of GPS devices for measurement of
movement patterns in confined spaces for court-based sports. J Sci Med Sport, 2010; 15: 523-525
Gray AJ, Shorter K, Cummins C, Murphy A, Waldron M. Modelling movement energetics using global
positioning system devices in contact team sports: limitations and solutions. Sports Med, 2018; 48: 1357-
1368
Hausswirth C, Le Meur Y, Couturier A, Bernard T, Brisswalter J. Accuracy and repeatability of the Polar1
RS800sd to evaluate stride rate and running speed. Int J Sport Med, 2009, 30: 354–359
Karaim ME, Aboelmagd N. GNSS Error sources, multifunctional operation and application of GPS. In: Rustam
B. Rustamov and Arif M. Hashimov, IntechOpen, 2018: DOI: 10.5772/intechopen.75493
Kluge F, Gaßner H, Hannink J, Pasluosta C, Klucken J, Eskofierm BM. Towards mobile gait analysis:
concurrent validity and test-retest reliability of an inertial measurement system for the assessment of
spatio-temporal gait parameters. Sensors, 2017; 17:1522. doi: 10.3390/s17071522
Malone JJ, Lovell R, Varley MC, Coutts AJ. Unpacking the black box: applications and considerations for using
GPS devices in sport. Int J Sports Physiol Perform, 2017; 12: 18-26
Mallo J, Mena E, Nevado F, Paredes V. Physical Demands of Top-Class Soccer Friendly Matches in Relation to
a Playing Position Using Global Positioning System Technology. J Hum Kinet, 2015, 47: 179-188.
McCall A, Nedelec M, Carling C, Le Gall F, Berthoin S, Dupont G. Reliability and sensitivity of a simple
isometric posterior lower limb muscle test in professional football players. J Sports Sci, 2015; 33: 1298-
304
Morrison K, Albert WJ, Kuruganti U. Biomechanical assessment of change of direction performance in male
university soccer players. 33rd International Conference on Biomechanics in Sports, Poitiers, France, June 29
- July 3, 2015
Nedelec M, McCall A, Carling C, Legall F, Berthoin S, Dupont G. The influence of soccer playing actions on
the recovery kinetics after a soccer match. J Strength Cond Res, 2014; 28: 1517–1523
O'Reilly M, Caulfield B, Ward T, Johnston W, Doherty C. Wearable Inertial Sensor systems for lower limb
exercise detection and evaluation: a systematic review. Sports Med, 2018; 48: 1221-1246
Roell M, Roecker K, Gehring D, Mahler H, Gollhofer A. Player monitoring in indoor team sports: concurrent
validity of inertial measurement units to quantify average and peak acceleration values. Front Physiol,
2018; 27: 141. doi: 10.3389/fphys.2018.00141
Takeda R, Lisco G, Fujisawa T, Gastaldi L, Tohyama H, Tadano S. Drift removal for improving the accuracy
of gait parameters using wearable sensor systems. Sensors, 2014; 14: 23230-47. doi: 10.3390/s141223230
Thornton HR, Nelson AR, Delaney JA, Serpiello FR, Duthie GM. Interunit reliability and effect of data-
processing methods of global positioning systems. Int J Sports Physiol Perform, 2019; 14: 432-438
van der Kruk E, Reijne MM. Accuracy of human motion capture systems for sport applications; state-of-the-
art review. Eur J Sport Sci, 2018; 18: 806-819
Varley MC, Fairweather I, Aughey, RJ. Validity and reliability of GPS for measuring instantaneous velocity
during acceleration, deceleration and constant motion. J Sport Sci, 2012; 30: 121-127
Varley MC, Jaspers A, Helsen WF, Malone JJ. Methodological considerations when quantifying high-intensity
efforts in team sport using Global Positioning System technology. Int J Sports Physiol Perform, 2017;
12:1059-1068
Vickery WM, Dascombe BJ, Baker JD, Higham DG, Spratford WA, Duffield R. Accuracy and reliability of GPS
devices for measurement of sports-specific movement patterns related to cricket, tennisand field-based
team sports. J Strength Cond Res, 2014; 28: 1697–1705
Vigh-Larsen JF, Dalgas Uandersen TB. Position-specific acceleration and deceleration profiles in elite youth
and senior soccer players. J Strength Cond Res, 2018; 32: 1114-1122
9
Waldron M, Murphy AA. comparison of physical abilities and match performance characteristics among elite
and sub-elite under-14 soccer players. Pediatr Exerc Sci, 2013; 25: 423-34
Welk G. Use of accelerometry-base activity monitors to assess physical activity. In: Welk G, editor. Physical
Activity Assessments for Health-Related Research. Champaign, IL: Human Kinetics; 2002. pp. 125–141
Yang S, Mohr C, Li Q. Ambulatory running speed estimation using an inertial sensor. Gait Posture, 2011; 34:
462-466
Zaferiou AM, Ojeda L, Cain SM, Vitali RV, Davidson SP, Stirling L, Perkins NC. Quantifying performance on
an outdoor agility drill using foot-mounted inertial measurement units. PLoS One, 2017; 12, e0188184:
doi: 10.1371/journal.pone.0188184
Figure 1. The SAFT
90
course performed by participants. The order of events is: Blue (forwards, backwards or
sidesteps), Red (Accelerate), Green (sidestep right-left), Red (accelerate, 180° turn, sprint). The course was
repeated 11 times at random and intermittent velocities.
Figure 2. The PlayerMaker™ system fitted to the ankle of the right foot. The complete system includes
an identically sized and fitted inertial measurement unit for the left ankle.
10
Figure 3. PlayerMaker™ gait tracking flowchart. 502 - accelerometer, 503 - gyroscope, 504 – "R" rotation matrix
(sensor relative to local frame), 506, 508, 510 – numeric integration of raw inputs, 512 - proprietery machine
learning for gait phase detection, 514 - proprietary Kalman Filter design to calculate position, velocity and
orientation. "+" sign indicates a sum, the "X" indicates a cross product. The rotation matrix transforms the
accelerations from the sensor frame to the local frame. At block 502, the processor receives sensor data from
the accelerometer. At block 503, the processor receives the sensor data from the gyroscope. At block 504 the
acceleration data rotated to the local frame and then subtracted by g on the local z-axis. At block 506, the
processed acceleration data are integrated and the velocity vector is formed at block 508. The velocity is
integrated and the position vector, along with the velocity vector, is used at block 514, with the Kalman Filter.
At block 510 the gyroscope data are integrated for calculation of R (that is used in block 504) and for the
detection of zero-velocity update and stance, along with the raw acceleration and gyroscope data, at block 512.
(Solid lines = feedforward to gait phase detection or Kalman Filter; dashed lines = feedback)
Figure 4. Distance covered in velocity zones of the PlayerMaker™ and three Global Positioning System (GPS)
devices. * = significantly (p < 0.05) different to all GPS devices. ƚ = significantly (p < 0.05) different to
PlayerMaker™.
11
Figure 5. Velocity profiles of the PlayerMaker™ and three GPS devices 60-s segment of the SAFT
90
protocol in a representative participant. COD = change of direction.
Figure 6. Velocity traces of two PlayerMaker™ systems during a 75-s segment of the SAFT
90
protocol
in a representative participant.
12
Table 1. Mean differences and 95% Limits of Agreement (95% LoA) for time motion analysis
variables between the PlayerMaker™ (inertial measurement unit) and Global Positioning System
#1 and #2 (GPS) devices during the SAFT90 protocol.
Movement variables PlayerMaker™ vs.
GPS#1
PlayerMaker™ vs.
GPS#2
Mean velocity (m/s) -0.17 ± 0.12* -0.09 ± 0.05*
Peak velocity (m/s) 0.03 ± 0.71 0.12 ± 0.41
Total distance (m) -30.08 ± 12.50* -32.15 ± 19.00*
Distance < 1.5 m/s (m) 29.14 ± 17.26* 18.99 ± 17.28*
Distance 1.5-3.5 m/s (m) -30.05 ± 32.45* -31.85 ± 33.05*
Distance 3.51-5.5 m/s (m) -29.06 ± 29.41* -21.39 ± 25.41*
Distance > 5.5 m/s (m) -0.11 ± 2.53 2.10 ± 4.51*
Peak acceleration (m/s2) -1.55 ± 1.61* -0.34 ± 1.02*
Acceleration count < 1.5 m/s2 49 ± 40* 29 ± 31*
Acceleration count 1.5-3.5 m/s2 -21 ± 12* 7 ± 15*
Acceleration count > 3.5 m/s2 -6 ± 5* -4 ± 6*
Peak deceleration (m/s2) -1.56 ± 1.38* -0.60 ± 1.33*
Deceleration count < 1.5 m/s2 195 ± 69* -26 ± 44*
Deceleration count 1.5-3.5 m/s2 -22 ± 17* 6 ± 14*
Deceleration count > 3.5 m/s2 -2 ± 3* -1 ± 3*
Mean acceleration distance (m) -0.02 ± 0.04* -0.02 ± 0.03*
Total acceleration distance (m) 10.92 ± 15.89* 8.22 ± 17.35*
Mean deceleration distance (m) -0.17 ± 0.04* -0.02 ± 0.03*
Total deceleration distance (m) -20.47 ± 11.83* -20.20 ± 17.44*
Note: * = significantly different (p < 0.05) to the corresponding PlayerMaker™ system during the
SAFT90. Minus value = lower than PlayerMaker™
13
Table 2. Mean differences and 95% Limits of Agreement (95% LoA) for time motion
analysis variables within the PlayerMaker™ (inertial measurement unit) and
compared to Global Positioning System #3 (GPS) device during the SAFT
90
protocol.
Movement variables PlayerMaker™ vs.
GPS#3
PlayerMaker™ vs.
PlayerMaker™
Mean velocity (m/s) -0.05 ± 0.03* 0.00 ± 0.00
Peak velocity (m/s) -0.09 ± 0.40 0.06 ± 0.25
Total distance (m) -16.78 ± 9.45* -0.16 ± 1.28
Distance < 1.5 m/s (m) 27.77 ± 7.16* 1.22 ± 2.79
Distance 1.5-3.5 m/s (m) -20.90 ± 26.47* -1.58 ± 4.46
Distance 3.51-5.5 m/s (m) -23.31 ± 27.15* 0.02 ± 2.38
Distance > 5.5 m/s (m) -0.35 ± 2.58 0.18 ± 0.81
Peak acceleration (m/s2) -1.75 ± 0.91* 0.04 ± 0.77
Acceleration count < 1.5 m/s2 130 ± 54* -3 ± 21
Acceleration count 1.5-3.5 m/s2 -29 ± 13* -1 ± 6
Acceleration count > 3.5 m/s2 -6 ± 5* 0 ± 1
Peak deceleration (m/s2) -1.56 ± 1.30* -0.09 ± 0.61
Deceleration count < 1.5 m/s2 111 ± 68* -1 ± 16
Deceleration count 1.5-3.5 m/s2 -27 ± 10* -3 ± 3
Deceleration count > 3.5 m/s2 -2 ± 3* 0 ± 1
Mean acceleration distance (m) -0.08 ± 0.03* 0.00 ± 0.01
Total acceleration distance (m) 11.77 ± 13.01* -0.56 ± 12.52
Mean deceleration distance (m) -0.09 ± 0.05* 0.00 ± 0.01
Total deceleration distance (m) -8.58 ± 9.42* -2.13 ± 5.98
Note: * = significantly different (p < 0.05) to the corresponding PlayerMaker™
system during the SAFT90. Minus value = lower than PlayerMaker™
