Article Review #1

A

c p m c 1 d v t o s ©

K

1

d o s t p m r s i a o b s

1 d

Available online at www.sciencedirect.com

Journal of Science and Medicine in Sport 13 (2010) 523–525

Original paper

Accuracy and reliability of GPS devices for measurement of movement patterns in confined spaces for court-based sports

Rob Duffield a,∗, Machar Reid b, John Baker c, Wayne Spratford c a School of Human Movement Studies, Charles Sturt University, Australia

b Tennis Australia, Australia c Australian Institute of Sport, Biomechanics & Performance Analysis Department, Australia

Received 18 February 2009; received in revised form 23 June 2009; accepted 21 July 2009

bstract

The aim of this study was to assess the accuracy and reliability of global positioning system (GPS) measures of distance and speed, ompared to a high-resolution motion analysis system, for confined movement patterns used in many court-based sports. A single male articipant performed 10 repetitions of four respective drills replicating court-based movement patterns and six repetitions of a random ovement drill that replicated tennis match-play movement patterns. Two 1 Hz and two 5 Hz GPS devices concurrently measured distance

overed and speed of all court-based drills. A 22 camera VICON motion analysis system, operating at 100 Hz, tracked the position of an 8 mm reflective marker affixed to one of the GPS devices to provide the criterion movement data. Results indicated that both 1 and 5 Hz GPS evices under reported distance covered as well as both mean and peak speed compared to the VICON system (P < 0.05). The coefficient of

ariation for both GPS devices for distance and speed measures ranged between 4 and 25%. Further, the faster the speed and more repetitive he movement pattern (over a similar location), the greater the measurement error. The inter-unit reliability for distance and speed measures f both 1 and 5 Hz systems for movements in confined spaces was generally low to moderate (r = 0.10–0.70). In conclusion, for court-based ports or movements in confined spaces, GPS technology under reports distance covered and both mean and peak speed of movement.

ights r

p i ( r o t m d a G p

2

2009 Sports Medicine Australia. Published by Elsevier Ltd. All r

eywords: Movement analysis; Team sports; Validity

. Introduction

Accurate assessment of the movement profile of athletes uring training and match-play can assist in the devel- pment of specific conditioning activities and recovery trategies. Accordingly, the use of global positioning sys- em (GPS) technology has been used to measure movement atterns, specifically distance covered and speed of move- ent, in many intermittent-sprint sports.1 Recent studies have

eported the reliability of GPS measures of distance and peed, and have also attempted to establish the criterion valid- ty of speed measures.2,3 However, to date, speed has been ssessed via the comparison of GPS data (collected at either 1

r 5 Hz) to infra-red timing systems that report a mean speed, ut not a peak or a continuous measure of the fluctuation in peed. Further, while previous literature has highlighted the

∗ Corresponding author. E-mail address: [email protected] (R. Duffield).

1 r a t

440-2440/$ – see front matter © 2009 Sports Medicine Australia. Published by El oi:10.1016/j.jsams.2009.07.003

eserved.

otential use of GPS technology for intermittent-sprint sports n open spaces and over prolonged continuous movements i.e. football codes and hockey)2–5; as yet the accuracy and eliability of GPS technology for use in court-based sports r movement patterns in confined spaces is unknown. Addi- ionally, no research has reported a true validation of GPS easures of speed against high resolution motion analysis

evices (i.e. VICON motion analysis system). Therefore the im of this study was to assess the accuracy and reliability of PS measures of distance and speed for confined movement atterns as used in court-based sports such as tennis.

. Methods

One moderately trained male (age: 30 years, height:

76 cm, mass: 70 kg) completed 10 repetitions of four espective drills replicating court-based movement patterns nd six repetitions of a random movement drill that replicated ennis match-play movement patterns. The four court-based

sevier Ltd. All rights reserved.

5 and M

m p s ( d e s o m b o s s o

r h w S G a a s c p G L m t a w (

p A i i s e a

a o h t e b e

3

e T o a t u m r

T M r

D

M

P

24 R. Duffield et al. / Journal of Science

ovement drills consisted of (1) a slow jog in a rectangle attern around the lines of the baseline, singles sideline and ervice line of a standard tennis court (26 m in total distance); 2) a fast run in a rectangle pattern as reported for the previous rill; (3) a side-to-side movement pattern approximately 2 m ither side of the centre line of a tennis court baseline and (4) a ide-to-side movement pattern approximately 4 m either side f the centre line of a tennis court baseline. The random move- ent drill consisted of 6 s of random movement around the

aseline of a tennis court replicating movements commonly bserved in tennis match-play. Each drill was timed with a topwatch and the subject maintained a stationary pose at the tart and end of each respective drill to assist determination f different drills for later analysis.

During each drill, GPS data were collected concur- ently by four devices housed in individual, customised arnesses worn simultaneously between the scapulae in line ith the lower cervical spine. Two 5 Hz (MinimaxX, Team port Model, Catapult, Australia) and two 1 Hz (SPI elite, PSports, Australia) GPS units were used in concert, with

ll devices activated 15-min prior to data collection to allow cquisition of satellite signals. Further, at all times, an ‘open’ ky was present and there were no obstructions, ensuring lear space for satellite acquisition. Speed was calculated ost hoc by customised software specific to each brand of PS unit (GPSports; Team AMS v2.1, MinimaxX; v28.5 ogan Software). Additionally, to determine criterion move- ent distance and speed data during the respective drills,

he three-dimensional (3D) position of a reflective marker, ttached to a harness of one of the participant’s GPS devices, as recorded by a 22 camera VICON motion analysis system

Oxford Metrics, UK) operating at 100 Hz. All testing was

a d d w

able 1 ean ± SD for VICON system, 1 and 5 Hz GPS devices for distance covered, mea

andom movement tennis drills.

VICON 5 Hz #1

istance (m) Slow 25.8 ± 1.5 23.9 ± 2.0 Fast 25.6 ± 0.3 21.3 ± 3.6a 2-m tennis 37.0 ± 0.7 34.2 ± 1.2a , c , d 4-m tennis 33.7 ± 0.5 34.6 ± 6.3a , c , d Random 12.0 ± 2.4 8.3 ± 1.7a

ean speed (m s−1) Slow 1.7 ± 0.1 1.6 ± 0.2 Fast 2.9 ± 0.2 2.4 ± 0.4a , c , d 2-m tennis 1.1 ± 0.1 1.0 ± 0.1a , c , d 4-m tennis 1.7 ± 0.1 1.8 ± 0.3c , d Random 2.0 ± 0.4 1.4 ± 0.3

eak speed (m s−1) Slow 2.6 ± 0.1 3.1 ± 0.7a , d Fast 4.9 ± 0.2 4.7 ± 1.6 2-m tennis 2.6 ± 0.2 2.7 ± 0.4d 4-m tennis 3.9 ± 0.2 4.9 ± 1.2a , d Random 3.9 ± 0.8 3.3 ± 1.5

a Within each respective drill type: significantly different to VICON. b Within each respective drill type: significantly different between 1 Hz models. c Within each respective drill type: significantly different between 5 Hz models. d Within each respective drill type: significantly different between 1 and 5 Hz mo

edicine in Sport 13 (2010) 523–525

erformed at night, outdoors on a plexicushion court at the ustralian Institute of Sport. Relevant static and dynamic cal-

bration was undertaken to accurately determine the 3D space n which the movement drills were completed. The VICON ystem was calibrated to an accuracy of less than 1 pixel for ach camera, with camera resolutions of 12 MP, representing n error of 0.0008%.6

Prior to statistical analyses, all VICON data was time ligned with the GPS units to match movement duration. A ne-way analysis of variance (ANOVA) with Tukeys’ post oc tests was used to analyse the difference between respec- ive measurement devices for both distance and speed within ach court-based drill. The inter-unit reliability was assessed y intra-class correlation (ICC) analysis, while the typical rror was expressed as a coefficient of variation (CV).

. Results

Seven satellites were acquired during the testing session, nsuring adequate satellite acquisition for data collection. able 1 shows the distance covered and mean and peak speeds f each court-based movement drill for each device. Both 1 nd 5 Hz devices showed significant differences (P < 0.05) o the VICON measures of distance and speed. In partic- lar, these differences were greater during the faster speed ovements, particularly when movement patterns involved

epeated movements over a similar location in space (2-m

nd 4-m drills). Table 2 shows the ICC and CV for each GPS evice for all movement drills. ICC values for both 1 and 5 Hz evices were low to moderate for most drills (0.10–0.70), hile CV’s for respective devices were between 4 and 25%.

n speed and peak speed of slow and fast rectangle drills and 2-m, 4-m and

5 Hz #2 1 Hz #1 1 Hz #2

23.6 ± 1.8a , d 25.5 ± 1.1 25.9 ± 0.7 18.0 ± 2.6a , c , d 20.9 ± 2.6a 21.8 ± 1.1a 30.7 ± 1.0a , c , d 23.1 ± 0.7a 23.7 ± 0.7a 28.5 ± 2.7a , c , d 24.0 ± 1.3a 22.8 ± 1.5a

8.3 ± 1.9a 8.0 ± 1.2a 7.9 ± 1.7a

1.6 ± 0.1 1.5 ± 0.1a 1.6 ± 0.1 2.1 ± 0.3a , d 2.0 ± 0.2a , b , c 2.6 ± 0.2a , b 0.9 ± 0.1a 0.8 ± 0.1a , d 0.9 ± 0.1a , d 1.5 ± 0.1a , c , d 1.4 ± 0.1a , b , c 1.1 ± 0.1a , b , c 1.4 ± 0.3 1.3 ± 0.2a 1.4 ± 0.6

3.0 ± 0.3d 2.3 ± 0.1d 2.4 ± 0.1d 3.9 ± 0.8a 3.7 ± 0.6a 4.2 ± 0.3 2.5 ± 0.4d 1.9 ± 0.1a , d 1.9 ± 0.1a , d 4.1 ± 0.8d 2.7 ± 0.2a , d 2.3 ± 0.4a , d 3.2 ± 0.9 2.4 ± 1.1a 2.5 ± 1.2a

dels.

R. Duffield et al. / Journal of Science and M

Table 2 Intra-class correlations (ICC) and co-efficient of variation (CV) within GPS device (between respective 1 Hz and 5 Hz models) for respective slow rect- angle, fast rectangle, 2-m and 4-m and random movement tennis drills.

5 Hz ICC 1 Hz ICC 5 Hz CV 1 Hz CV

Distance (m) Slow 0.38 0.17 9.8 3.6 Fast −0.06 0.20 17.8 9.5 2-m tennis 0.01 0.24 3.5 3.6 4-m tennis 0.15 0.15 11.0 5.8 Random 0.43 0.87 16.8 7.6

Mean speed (m s−1) Slow −0.09 0.45 9.1 2.1 Fast 0.05 0.25 17.1 11.1 2-m tennis 0.41 0.70 3.4 3.9 4-m tennis 0.03 0.55 15.6 5.6 Random 0.43 0.69 16.9 19.3

Peak speed (m s−1) Slow 0.06 0.70 17.6 2.3

4

a s c n G c d i s t p d m a C m c t r

s s a ( t g o i t u

d

5 a c b i t p t f f t u 5 d d m a

m m d t t F t a s o d

P

•

•

R

global positioning system for assessing player movement patterns in field

Fast −0.12 −0.08 31.7 15.3 2-m tennis 0.69 0.22 20.3 5.8 4-m tennis 0.08 0.11 24.5 12.6 Random −0.08 0.93 35.3 26.7

. Discussion

Estimation of the physical load associated with training nd matchplay is a source of interest for many court-based ports. However, questions pertaining to the validity of the alculated measures may limit the integration of this tech- ology. Previous research on the accuracy and reliability of PS devices for measuring athlete movement patterns indi-

ates that for slow to moderate speeds over moderate to longer istances, GPS technology has acceptable inter-unit reliabil- ty and accuracy for distance measured.2–5 However, as the peed of movement increases, the reliability of 1 Hz GPS echnology has been reported as somewhat lower.2 Further, revious research has outlined the error measurement of GPS evices to range between 2 and 25%, depending on move- ent distance and speed.2–5 The present research generally

grees with this previously published data, with inter-unit V for distance covered ranging from 4% for slower move- ent speed, up to 30% for faster movement speed. Moreover,

ompared to the VICON system, GPS devices underestimate he distance covered, particularly at higher speed and during epeated movement patterns in confined spaces.

The current study is the first to have a true criterion mea- ure of speed through the use of a VICON motion analysis ystem capturing data at 100 Hz. Results indicated that both 1 nd 5 Hz systems underestimated both peak and mean speed by 10–30%) during court-based movement drills. Again, he faster movement speeds result in an increased CV, sug- esting that as compared to VICON, the lower resolution f GPS devices fails to capture all relevant data. Accord- ngly, during movements performed at high velocity and/or

hat include frequent changes of direction, this manifests to nder-represent mean and peak velocity measures.

While differences between respective 1 and 5 Hz GPS evices are expected due to the resolution of the units, a

edicine in Sport 13 (2010) 523–525 525

Hz device seemed more accurate when measuring distance nd velocity for movement patterns at higher velocities in onfined spaces. However, there was still some discrepancy etween 5 Hz models and VICON measures. Further, the nter-unit reliability for the 5 Hz system seemed to be lower han that of the 1 Hz model. Given many of the movement atterns in small areas or for court-based sports involve repe- ition of movement in the same location in space, the greater requency of measurement in the 5 Hz system may account or the generally more accurate measures of speed noted in he 5 Hz devices. Despite greater accuracy, and while spec- lative, the observed reduction in inter-unit reliability in the Hz model may relate to the increased volume of collected ata in the 5 Hz model, or differences in treatment of acquired ata by the respective systems. Regardless, in sports where easures of mean and peak speed are of interest, it appears

dvisable to use higher resolution devices. In conclusion, the use of the current GPS technology to

easure distance and speed of court-based or confined space ovements likely leads to an underestimation of movement

istance and speed. Consequently, practitioners using this echnology to monitor athlete activity should account for his underestimation when monitoring movement patterns. urther, due to lower inter-unit reliability, it is advisable

hat participants and athletes wear the same device within ny research design or while training in the field. Finally, it hould be noted that these results apply to current versions f GPS software and hardware, which continue to undergo evelopment.

ractical implications

GPS devices underestimate the distance covered and speeds recorded during court-based, confined movement patterns. Where possible, the same GPS device should be used for the same player or subject in order to avoid inter-unit error.

eferences

1. Reid M, Duffield R, Dawson B, et al. Quantification of the physiological and performance characteristics of on-court tennis drills. Br J Sports Med 2008;42:146–51.

2. Coutts AJ, Duffield R. Validity and reliability of GPS devices for mea- suring movement demands of team sports. J Sports Sci Med 2010;13: 133–5.

3. Portas M, Rush C, Barnes C, et al. Method comparison of linear distance and velocity measurements with global positioning satellite (GPS) and timing gates. J Sports Sci Med 2007;6(Suppl. 10):7–8.

4. Edgecomb SJ, Norton KI. Comparison of global positioning and computer-based tracking systems for measuring player movement dis- tance during Australian football. J Sci Med Sport 2006;9(1/2):25–32.

5. MacLeod H, Morris J, Nevill A, et al. The validity of a non-differential

hockey. J Sports Sci 2009;27:121–8. 6. Elliott B, Adlerson J. Laboratory versus field testing in cricket bowling:

a review of current and past practice in modelling techniques. Sports Biomech 2007;6:99–108.