Jour nal of Sports Sciences, 1999 , 17, 249± 257

M easuring running speed using photocells

1 2 1
M .R. YEADO N, * T. K ATO and D.G. K ERW IN
1
Department of Sport Science, Physical E ducatio n and Recreation M anagem ent, Loughborough U niversity, A shby Road,
2
Loughborough LE 11 3TU , U K and Laborator y for Exercise Physiology and B iom echanics, Chukyo U niversity,
Tokodate 101, K aizucyo, Toyota 470-03, Japan

Accepted 7 June 1998

Photocell tim ing systems are used routinely to measure running speeds. In this study, the accuracy of such
systems was evaluated using centre of m ass speed estim ates from three-dimensional video analysis as criteria.
One subject ran at W ve nominal speeds (5± 9 m ´ s - 1 ) for each of W ve separations (1.6± 2.4 m ) between consecutive
photocells. Running speeds were calculated from the photocell data using single beam and double beam system s.
For single beam systems, the start of the W rst break of a beam and the start of the longest break of a beam were
used as trigger criteria. For double beam system s, the W rst occurrence of both beams being broken and the start
of the longest double break were used as trigger criteria. Root mean square speed errors were smaller for the
double beam system s. The longest break criterion gave smaller root m ean square errors than the W rst break
criterion. In general, errors in speed were smaller for greater photocell separations. An error of 0.1 m ´ s- 1 was
achieved using a single beam system set at hip height with a longest break criterion for photocell separations of
around two stride lengths. The advantage of using a double beam system is that it achieves this accuracy without
the need to adjust photocell separation for diV erent stride lengths.

K eyw ords : beams, photocells, running, speed, speedometer, timing.

Introduction m otion of an athlete is to be recorded using this

Photocells have been used for m easuring speeds in
various sporting activities, including golf, javelin,
soccer, luge (Richards et al., 1985; Viitasalo and Korjus,
1987; Poulm edis et al., 1988; Br  ggem ann and Kunkel,
1990) and running (Ikai, 1968; M ero and K om i, 1985;
Ae et al., 1985; N igg et al., 1987). Although photocells
are used routinely to estim ate running speed, little
inform ation exists on the accuracy of such system s.
T he beam from a photocell emitter goes to a reX ector,
located directly opposite, and is reX ected back to the
photocell sensor where it is detected. If sets of em itter± -
reX ector pairs are arranged at equal separations along
a runway, a series of beam s is produced that can be
successively interrupted by an object passing through
the beam s. If the voltage output from each of the beam s
is logged, a pulse wave will be produced, w hich rises on
beam interruption and falls when the beam is reinstated.
K now ing the beam separations and the tim e history of
the beam breaks it is possible to calculate the average
speed of the object between any two beam s. If the
* Author to w hom all correspondence should be addressed.
technique, it is W rst necessary to adopt a deW nition of
running speed.
Running speed will be deW ned as the m ean speed of
the m ass centre over a given interval. Even if som eone
is running at `constant speed’ , the speed of the m ass
centre w ill vary during a stride, since the body de-
celerates and then accelerates with each foot contact.
Since a photocell beam is broken by the leading edge
of part of the body rather than by the m ass centre, the
speed estim ate obtained using photocells will never
correspond exactly to the m ass centre speed. N everthe-
less, if each beam is broken by the sam e part of the body
and the body is in the sam e conWguration as each beam
is broken, then the correspondence between speed
estim ates is likely to be ver y close.
If a subject breaks one beam w ith an arm and the next
w ith the torso, a considerable error could be introduced
into the speed estim ate, since the torso m ight be 20± 30
cm behind the leading arm . T his problem m ight be
reduced by placing the photocell beam s at hip height
so that the arm s do not break beam s. In this case, there
still m ight be a problem of breaking the beam w ith a
leading thigh. Another approach could be to use the

0264 ± 0414/99 Ó 1999 E & FN Spon

250
start of the longest photocell break as a trigger criterion.
As the torso will produce a longer break than an arm ,
this m ethod could rem ove arm triggering; a leading
thigh m ight still, on occasion, be the trigger. Using a
double beam system could also be a possible solution to
this problem . Two photocells could be set at diV erent
heights and be required to be broken sim ultaneously.
This would reduce the possibility of triggering the
system with an arm or a leg and would increase the
possibility of selecting the instant that the torso breaks
the beam s.
T he distance between consecutive photocells m ight
be im portant. E ven if the W rst beam and the second
beam were broken by the sam e part of the body, for
exam ple the torso, the inclination of the torso m ight be
diV erent. If the distance between consecutive photocells
was set to a m ultiple of the athlete’ s stride length, the
inclinations of the torso would be sim ilar and the speed
obtained would be m ore likely to be an accurate
estim ate of the m ass centre speed.
Ideally, an exact m easure of running speed is required
to evaluate the accuracy with w hich photocell system s
can m easure running speed. It is suY cient, however,
to use a m easurem ent system that has accuracy of a
higher order. O ne possibility for such a system is to
m easure running speed using W lm or video analysis
together with a linked segm ent m odel of the hum an
body (H ay and M iller, 1985; M ero and K om i, 1985;
Hay et al., 1986; Ham ilton, 1993). W ith such an
approach, each segm ent is assum ed to be rigid and to
share a com m on point with each adjacent segm ent. If
the location of the m ass centre of each segm ent, relative
to digitized points on each segm ent, is known and the
relative segm ental m asses are known, the location of the
whole body m ass centre can be calculated. D em pster
(1955) repor ted segm ental inertia values obtained
from eight cadavers and these data are com m only used
with linked segm ent m odels. T he m ass distribution of
individual athletes, however, will diV er from average
cadaver data. To obtain personalized segm ental inertia
param eters from anthropom etric m easurem ents on an
individual, m odels of the body based upon geom etric
solids have been developed (Jensen, 1978; H atze, 1980;
Yeadon, 1990).
T he purpose of this study was to evaluate the accur-
acy of a photocell system in estim ating running speeds
using speeds from segm ental video analysis as criteria.
Our speciW c aim s were to: (1) com pare the results of
single beam system s m ounted at diV erent heights;
(2) com pare the diV erence between using the W rst break
and the longest break as the trigger criterion; (3) com -
pare the results of a single beam system and a double
beam system ; and (4) determ ine the relationship
between the photocell separation and the accuracy of
speeds from the photocell system .
Yeadon et al.
M ethods
Data collection
One healthy and physically W t m ale athlete took part
in the study. His height and body m ass were 190 cm
and 80.7 kg, respectively. N inety-W ve anthropom etric
m easurem ents were taken using anthropom etric cali-
pers and a m easuring tape, including 34 lengths, 41
perim eters, 17 widths and 3 depths (Yeadon, 1990).
These m easurem ents were required to locate the w hole-
body m ass centre during running.
T he subject ran down a running lane deWned by
m arkings on the wooden X oor of a sports hall. Two
gen-locked video cam eras, operating at 50 W elds per
second, were used to record running perform ance in all
trials. The Panasonic F-15 (sVH S) cam era and the Sony
Handycam Pro (Hi-8) cam corder had their cam era axes
angled to cover the sam e horizontal W eld of view of 10 m
and were placed at a distance of 14 m from the m id-line
of the running lane with the distance between the two
cam eras set at 2.40 m (Fig. 1). T he shutter speeds of
these cam eras were 1/250 s for the Panasonic cam era
and 1/300 s for the Sony cam era.
To obtain data for calibrating the cam eras using a
direct linear transform ation (DLT) procedure, 12 cali-
bration poles, each with three m arkers located at heights
of 0.13, 1.13 and 2.13 m , were placed in the W eld of
view of each cam era. T hese calibration poles were posi-
tioned as shown in F ig. 1 and their locations surveyed.
The surveying equipm ent consisted of an electronic
theodolite, w hich m easured horizontal and vertical
angles, and an electrom agnetic distance m easurem ent
device, w hich determ ined distance by m easuring the
phase diV erence between transm itted and reX ected
electrom agnetic waves. Three control stations were used
and the control network was calibrated by m easure-
m ent of distances and angles to two stations from the
third. T his was repeated at all three stations. The centres
of the 36 m arkers on the poles were surveyed using
the theodolite to obtain two angles for each m arker
from each station. T he three-dim ensional locations of
the m arkers were obtained using a least-squares solution
to the intersection of a ray from each control station.
The use of three control stations enabled error estim ates
of each of the m arker coordinates to be obtained.
Two cam eras were used to perm it a three-dim ensional
analysis, since a planar analysis is prone to parallax
or perspective error w hen points lie outside the cali-
bration plane (Doolittle, 1971). W ith a two-dim ensional
analysis using a single cam era 14 m from the calibration
plane, a system atic lateral deviation of the runner by
0.14 m towards the cam era would result in a 1% over-
estim ation in the calculated speed of the m ass centre.
The two cam eras were placed close together to increase
M easuring running speed
F igu re 1 Location of experimental equipm ent. The photo-
cell separation d varied from 1.6 to 2.4 m .
the accuracy of displacement estim ates in the direction
of the run at the expense of the accuracy in the lateral
direction. T he m ass centre m ean speed over an inter val
was calculated as the displacem ent in the W xed direction
of the run divided by the tim e.
Six tripods were used to m ount three pairs of photo-
cells and corresponding reX ectors. The lower photocells
were m ounted at a height of 1.05 m , around the level of
the hips, and the upper photocells were m ounted 0.20 m
higher. The distance between the photocells and the
reX ectors was 1.5 m , while the separation of adjacent
photocells was varied in increm ents of 0.2 m . The
photocell sensor detected w hen the beam was broken by
a change in the analog signal, w hich was captured by an
analog-to-digital converter (C ED 1401) connected to
an Archim edes 410 m icrocom puter. T he athlete was
given feedback on his speed of running during a period
of fam iliarization w ith the set-up. He was then asked to
run at W ve diV erent nom inal speeds (5, 6, 7, 8 and 9
m ´ s - ) for W ve diV erent photocell separations (1.6, 1.8,
1
2.0, 2.2 and 2.4 m ) show n as d in Fig. 1, giving a total
of 25 trials.
D ata processing
All video im ages were digitized using an Apex Im ager
24 bit colour im age capture board in a Archim edes 410
m icrocom puter running Target software. A Panasonic
AG -7350 video-cassette recorder and a high-resolution
Sony PO M -1444Q M colour video m onitor were used.
T here are 768 pixels horizontally and 576 pixels
vertically in the Apex Im ager video fram e store. The
Target video analysis system used a cursor with sub-
251
pixel m ovem ent that gives a m easurem ent resolution of
6144 horizontally and 4608 vertically (Kerw in, 1995).
T he im ages of the 36 calibration m arkers were digitized
in W ve consecutive W elds for each cam era. T hese digit-
ized coordinates, together with the surveyed three-
dim ensional locations of the 36 points, were used to
calculate 12 cam era param eters for each cam era using
a D LT procedure w ith a correction for radial lens
distortion (K arara, 1980). An unbiased estim ate of
reconstruction accuracy was obtained by calculating the
root m ean square diV erence between the surveyed
coordinates and the reconstructed coordinates by
m aking allowance for the num ber of degrees of freedom
as described in Yeadon and M orlock (1989).
For each running trial and each cam era, every video
W eld was digitized for three strides starting Wve W elds
before the beginning of the W rst foot contact and ending
W ve Welds after the end of the four th foot contact.
F ifteen points on the body were digitized: wrist, elbow,
shoulder, hip, knee and ankle centres together w ith the
ends of the toes and the m iddle of the head. Joint centres
were identiW ed from a knowledge of anatomy together
w ith an assum ption that centres lay on the m id-lines of
lim b segm ent im ages. For video W elds in w hich body
landm arks were obscured by other parts of the body,
the `Target’ replay facility, which allows adjoining video
W elds to be viewed in sequence, was used to help
estim ate landm ark locations. T he three-dim ensional
coordinates of the body points were obtained using the
D LT reconstruction.
T he body segm ents of the m odel were sectioned into
40 solids w hose physical dim ensions were calculated
from the anthropom etric m easurem ents of the subject.
F rom these dim ensions and the segm ental densities
of Chandler et al. (1975), the m asses and m ass centre
locations of 14 body segm ents were calculated (Yeadon,
1990). T he relative segm ental m asses and locations of
segm ental m ass centres relative to the digitized body
points were then calculated and used, together with the
three-dim ensional coordinates of the digitized body
points, to obtain the location of the whole-body m ass
centre.
Stride length and m ass centre average speed over the
photocell separation were calculated from the un-
sm oothed coordinate data. T he m ean foot location
during ground contact was deW ned as the average
coordinate in the direction of the run of the ankle and
toe during the tim e that both heel and toe were in
contact with the ground. Stride length was deW ned as
the diV erence between the m ean foot location during
one foot contact and the m ean foot location for the next
(contralateral) foot contact. T he X at foot contact ranged
from three W elds (0.06 s) to Wve W elds (0.10 s), so that
stride length was calculated using the average of 6± 10
coordinate values.
252
Figure 2 Beam break durations from photocell system. a 1
and a 2 are the durations for the upper photocells W rst break for
the W rst and second intervals, respectively. b 1 and b 2 , c 1 and c 2
are the durations for upper photocells longest break and both
photocells simultaneous break, respectively.
T he m idpoint of consecutive foot locations was used
to deW ne a m id-stride location for each of the three
stride lengths for each trial. The three stride lengths and
associated m id-stride locations deWned stride length
as a piecewise linear function. Stride lengths were then
interpolated using this function for the two m id-
photocell locations. This was done to give a stride length
corresponding to each of the two photocell intervals.
T he m ass centre locations were known at inter vals
of 0.02 s; these were used to exp ress tim e as a piece-
wise linear function of m ass centre displacem ent. This
function was used to identify the tim es corresponding
to the three photocell displacem ents. T he m ass centre
average speed over one photocell separation was
calculated as the photocell separation divided by the
diV erence in the tim es corresponding to the two photo-
cell locations. As there were 25 running trials and
two photocell separations for each trial, a total of 50
criterion speeds were obtained.
Speeds were calculated from the photocell tim ings in
six diV erent ways, using the tim es from the start of: (1)
upper photocells W rst break, (2) lower photocells W rst
break, (3) upper photocells longest break, (4) lower
photocells longest break, (5) both photocells double
break, and (6) both photocells longest double break.
The tim e histories of the voltage output of three pairs
of photocells are represented in Fig. 2. W hen a beam
is broken, there is a rapid change in voltage; w hen
the beam is re-established, the voltage returns to the
original value. Figure 2 shows the procedure for cal-
culating the beam breaking durations for the upper
photocells W rst break (a 1 and a 2 ), the upper photocells
longest break (b 1 and b 2 ) and both photocells double
break (c 1 and c 2 ) using the upper and lower beam break
tim es at each of the three photocell stations.
Yeadon et al.
Data analysis
To estim ate the errors in the speed estim ates obtained
from video, the sources of error were identiW ed as the
calibration procedure, the digitizing process and the
estim ation of segm ental inertia param eters. To obtain
an independent set of estim ates of speed and stride
length from the video sequences, each of these three
procedures was repeated independently. T he calibration
digitization was carried out by an independent operator.
Five trials that covered the range of photocell separ-
ations and nom inal speeds were also redigitized by this
second operator. Two additional inertia sets based on
anthropom etric m easurem ents of a diver and a gym nast,
together w ith three additional inertia sets based on
m ean values from studies by D em pster (1955), C lauser
et al. (1969) and Z atsiorsky et al. (1990), were also
used to give W ve independent segm ental inertia sets.
A diV erent iner tia set was associated w ith each of the
Wve redigitized trials and the redigitized calibration
param eters were used to reconstruct the body land-
m arks to provide new speed and stride length estim ates.
The root m ean square diV erences between the original
speeds and the new speeds and between the original
stride lengths and new stride lengths were calculated.
These root m ean square values gave a m easure of the
reliability with which m ass centre speeds and stride
lengths could be calculated from digitized video data.
The errors between the calculated and the actual speeds
arise from both system atic and random errors in the
calibration and digitization processes and in the esti-
m ation of segm ental inertia param eters. Reliability
m easures typically reX ect the eV ects of random errors
in each process but do not take account of system atic
errors. T he error analysis procedure described above
was designed to include the eV ects of system atic errors
as far as possible. Subject to the assum ption that no
system atic errors have been excluded, the root m ean
square values m ay also be used as estim ates of accuracy.
Each of the six photocell speed data sets was
regressed against the digitized speed data set. The inter-
cepts and slopes were checked for signiW cant diV erences
from 0 and 1, respectively, to determ ine whether any
system atic errors were present in any of the six photocell
m easuring system s. T he deviations from the digitized
speeds were determ ined for each of the six photocell
speed data sets. T he m ean, standard deviation and root
m ean square were calculated for each of the six sets of
deviations.
To test for diV erences between the upper, lower and
double photocell system s, and between the W rst and
longest break criteria, the deviations that changed under
a change of condition were subjected to a variance ratio
F-test. The ratio of the variances was calculated as the
square of the ratio of the root m ean square deviations.
M easuring running speed
In addition, a binom ial sign test, based on the num ber
of im provem ents in those deviations that changed,
was used to test for signiW cant diV erences between
conditions.
Speeds were also calculated for each of the six con-
ditions over the double separation between the W rst and
third photocell stations, to test whether increasing the
photocell separation resulted in sm aller deviations.
Variance ratio tests and sign tests were again used to test
for signiW cant diV erences between single and double
separations.
To test whether deviations were sm aller for photo-
cell separations near to one stride length, the ratio of
photocell separation to stride length was determ ined
for each of the 50 speeds in each of the six conditions.
To rem ove the eV ect of the m agnitude of separation
on the deviations, the deviations were norm alized to a
set separation of 2.0 m based on the assum ption that
deviations were inversely proportional to separation.
For each of the six conditions, each data set ± com -
prising 50 speeds and stride lengths with associated
photocell separations ± was separated into inner and
outer data sets of equal size with the ratio of photocell
separation to stride length near to 1.0 for the inner set
and far from 1.0 for the outer set. The inner and outer
data sets were chosen so that the m ean separations
for each set were nearly equal and did not therefore
produce a system atic bias in the resulting deviations. An
F-test was used to com pare the ratio of the variances
of inner and outer speed deviations to test w hether
they were signiW cantly diV erent. The sam e procedure
was followed for the speeds calculated over a double
separation with a distance of 4 m used for norm alizing
the data.
R esults
E rror estimates
T he estim ated root m ean square errors in the surveyed
coordinates of the calibration points were less than
0.001 m in each direction. T he D LT reconstruction
error estim ates for the calibration points were 0.003 m
in the direction of running, 0.014 m laterally and 0.003
m vertically. T his reX ects the intention to increase the
accuracy of the displacement estim ates in the direction
of running at the exp ense of the accuracy in a lateral
direction. T he root m ean square diV erence between the
10 speeds obtained from the original video digitization
and the 10 speeds obtained from the redigitization was
0.037 m ´ s - 1. T he root m ean square diV erence between
the stride length values was 0.0066 m . T hese diV erences
reX ect the errors in both digitizations and, because
the variance of the diV erence is the sum of the two
253
Table 1 Average speeds and stride lengths
- 1
Nominal speed (m ´ s )
5 6 7 8 9
- 1
Running speed (m ´ s ) 5.16 5.87 6.91 7.79 8.50
Stride length (m) 1.81 1.95 2.09 2.17 2.09
Note: Stride length is the distance from one foot contact to the next
contact of the contralateral foot.
variances, the errors associated with a single digitization
m ay be obtained by dividing these values by Ö 2. T his
gives error estim ates of 0.026 m ´ s - 1 for speeds and
0.0047 m for stride lengths. As personalized estim ates
of inertia were used in the calculation of the criterion
speeds, it m ight be expected that the root m ean square
error in the criterion speeds would be less than 0.026
m ´ s- 1 .
Average stride length and running speed
T he m ass centre average speed over all trials at each
nom inal running speed and the corresponding average
stride length are presented in Table 1. The m ean
speeds were within 3% of the nom inal speeds with the
exception of the fastest, corresponding to a nom inal
speed of 9 m ´ s - 1 for which the m ean was 8.5 m ´ s - 1 .
T his m ay have been because the athlete needed to
slow down after passing through the photocell beam s
to avoid running into the end wall of the sports hall.
T he stride lengths tended to be greater for the faster
speeds.
Root mean square deviations in speed
For each of the regressions of the six sets of photocell
speeds against the criterion digitized speeds, no inter-
cept was signiWcantly diV erent from 0.0 (P > 0.10) and
no slope was signiW cantly diV erent from 1.0 (P > 0.30).
F igure 3 com pares the estim ates obtained from video
and photocells using the longest double break criterion.
Table 2 presents the m ean, standard deviation and root
m ean square of the deviations of the speeds calculated
using the photocell tim es from the criterion speeds
obtained from the video analysis. N one of the m ean
deviations show n were signiW cantly diV erent from zero
(P > 0.05). T he sum m ar y data are given for each of
the six calculation m ethods for speeds based on the
photocell break tim es. T he upper photocells W rst break
gave the largest root m ean square deviation in speed
(0.49 m ´ s - 1 ) and both photocells longest double break
gave the sm allest root m ean square deviation in speed
(0.18 m ´ s - 1).
254 Yeadon et al.

Table 3 com pares the speed deviations for the six
conditions using the variance ratio F-test and the bi-
nom ial sign test for those deviations that were altered
under a change of condition. The placem ent height of a
single beam system was im portant, as the lower beam s
gave deviations 1.6 tim es sm aller than the upper beam s
(P = 0.01). The use of a double beam system changed
10 of the speeds from the lower beam speed, with
im provem ent in nine cases reducing the root m ean
square deviation by a factor of 1.7 (P = 0.05). The use of
the longest break criterion did not im prove the upper
beam speeds and only changed two lower beam speeds,
although the m agnitude of these changes was large
enough to be signiWcant at P = 0.01 using the F-test.
The longest break criterion only changed one of the
double beam speeds, reducing the corresponding devi-
ation from 0.42 to 0.05 m ´ s - 1.
T he eV ect of doubling the photocell separation is
show n in Table 4. The root m ean square values were all

- 1
Table 2 Deviations of photocell speeds (m ´ s ) from
criterion video speeds

Trigger condition

Longest Longest Longest

Upper upper Lower lower Double double

Mean - 0.05 0.02 0.02 0.00 0.03 0.02

Figure 3 Regression of speed estimates obtained from s 0.49 0.49 0.30 0.27 0.19 0.18
photocells using the longest double break criterion and from RMS 0.49 0.49 0.30 0.27 0.19 0.18
video digitization. The intercept is not signiW cantly diV erent
from 0 and the slope is not signiW cantly diV erent from 1. Note: RM S = root m ean square.

- 1
Table 3 Com parison of root mean square (RM S) deviations in speed (m ´ s ) across trigger
conditions

Number of Number of Binomial RMS 1

Trigger condition changes improvements sign test RMS 2 F -test

Upper : longest upper 35 15 P > 0.05 1.02 P > 0.05

Upper : lower 47 33 P < 0.01 1.62 P < 0.01
Lower : longest lower 2 2 P > 0.05 15.87 P < 0.01
Longest upper : longest lower 48 38 P < 0.01 1.83 P < 0.01
Lower : double 10 9 P < 0.01 1.73 P < 0.05
Longest lower : longest double 8 8 P < 0.01 1.61 P > 0.05
Double : longest double 1 1 P > 0.05 8.71 P > 0.05

Table 4 Deviations in speed (m ´ s - 1) for single and double photocell separations

Single separation Double separation

Trigger RM S 1
condition M ean ± s RM S M ean ± s RM S RMS 2 F -test

Upper 0.05 ± 0.49 0.49 - 0.07 ± 0.24 0.25 1.95 P < 0.01
Lower 0.02 ± 0.30 0.30 0.01 ± 0.19 0.19 1.60 P < 0.01
Longest lower 0.00 ± 0.27 0.27 - 0.01 ± 0.14 0.14 1.90 P < 0.01
Longest double 0.02 ± 0.18 0.18 0.01 ± 0.07 0.07 2.78 P < 0.01

Note: The mean single separation was 2.0 m and the m ean double separation was 4.0 m . RM S = root
m ean square.
M easuring running speed 255

close to the standard deviation estim ates, since the m ean
deviations were all close to zero. As a consequence,
the root m ean square m ay be used as an estim ate of
the square root of the variance. For each of the upper,
lower, longest lower and longest double conditions,
the root m ean square deviations were reduced by 1.6± -
2.8 tim es (P = 0.01), with an overall average for the
four conditions of 2.06 tim es. T his is in line with the
theoretical expectation that a doubling of the separation
should result in a halving of the deviation in the cal-
culated speed. If the tim ing errors are assum ed to be
independent of photocell separation, a doubling of the
photocell separation will correspond to a doubling of
the tim e interval so that the relative tim ing error would
be halved. Figure 4 shows the distributions of absolute
deviations as a function of photocell separation for the
conditions upper (Fig. 4a), lower (Fig. 4b) and longest
double (Fig. 4c).
T he eV ect of the ratio of photocell separation to stride
length on the speed deviations is shown in Table 5. The
deviations for separations of around one stride length
were 1.2± 1.5 tim es sm aller than for separations away
from one stride length, although this was only signiW -
cant (P = 0.05) for the upper beam condition. It should
be noted that the m ean separations in the two cases
were both close to one stride length and did not there-
fore aV ect the deviations. T he variance of the outer
separations was, of course, m uch larger than the inner
separations. T he sam e com parison was m ade for
speeds calculated over a double separation w ith the
data being split according to w hether or not the separ-
ation was near to two stride lengths. T he deviations for
separations near to two stride lengths were 1.3± 1.8
tim es sm aller for the upper, lower and longest lower
conditions and 1.1 tim es larger for the longest double
condition, although the change was only signiW cant
(P = 0.05) for the longest lower beam condition (Table
6). This provides som e evidence that speeds calculated
for photocell separations close to a m ultiple of stride
length m ay be m ore accurate.

D iscussio n

T he error in the criterion speeds ± arising from errors

in the three-dim ensional D LT calibration, errors associ-
F igu re 4 Absolute speed error and photocell separation for ated with digitizing the body landm arks and errors
(a) upper photocells W rst break, (b) lower photocells W rst break arising from the particular inertia data set used ± was
and (c) both photocells longest simultaneous break. estim ated to be less than 0.026 m ´ s - 1 . Since the sm allest

Table 5 Deviations in speed (m ´ s- 1) for photocell separations close to one stride length

Not close to one stride length Close to one stride length

RM S 1
Trigger condition M ean ± s RMS M ean ± s RMS RM S 2 F -test

Upper 0.03 ± 0.54 0.52 - 0.13 ± 0.43 0.44 1.19 P > 0.05
Lower 0.06 ± 0.34 0.34 - 0.04 ± 0.23 0.23 1.47 P < 0.05
Longest lower 0.04 ± 0.32 0.32 - 0.04 ± 0.23 0.23 1.37 P > 0.05
Longest double 0.05 ± 0.19 0.19 - 0.02 ± 0.17 0.17 1.16 P > 0.05

Note: The m ean separation close to one stride length was 0.99 stride lengths (s = 0.07); the mean separation not close to one stride length was
0.99 stride lengths (s = 0.20). RM S = root mean square.
256
- 1
Table 6 Deviations in speed (m ´ s ) for photocell separations close to two stride lengths
Not close to two stride lengths
Trigger condition M ean ± s RM S
Upper - 0.02 ± 0.29 0.28
Lower 0.04 ± 0.23 0.23
Longest lower - 0.01 ± 0.18 0.18
Longest double 0.02 ± 0.06 0.06
Note: The m ean separation close to two stride lengths was 1.98 stride lengths (s = 0.15); the m ean separation not close to two stride lengths was
2.00 stride lengths (s = 0.39). RM S = root m ean square.
root m ean square deviations of the photocell speeds
from these criterion speeds were greater than 0.06
m ´ s - , these deviations m ay be considered to be reason-
1
able estim ates of the errors in the photocell speeds.
T he lower photocells were m ounted at hip height
(1.05 m ) and the upper photocells were m ounted 0.20
m higher. At the upper photocells, the beam s were
broken twice (separately by the arm s and the torso) in
60% of all cases. In contrast, at the lower photocells,
beam s were broken tw ice in just 4% of all cases. This
was reX ected in the root m ean square speed errors for
the two m ethods (0.49 m ´ s - for the upper beam s and
1
- 1
0.30 m ´ s for the lower beam s). For a single beam
system , the photocells should be m ounted at a height at
which only one part of the body usually breaks the beam ,
such as hip height as used in this study or possibly head
height (D yas and K erwin, 1995).
For photocells m ounted at hip height, the use of the
longest break criterion reduced two of the 50 speed
errors from over 0.5 m ´ s - to less than 0.1 m ´ s - . Even
1 1
so, the root m ean square speed error of 0.27 m ´ s - over
1
intervals ranging from 1.6 to 2.4 m is unacceptably high
for m ost purposes. For separations close to one stride
length, the root m ean square error of the longest break
lower photocell speeds was 0.23 m ´ s - , which is com -
1
- 1
parable with that of 0.17 m ´ s for the longest break
double beam speeds. It can be concluded that speeds
calculated from photocell tim es for separations of the
order of 2 m w ill be in error by 0.2 m ´ s - or m ore. If
1
speeds over individual strides are required, then three-
dim ensional W lm or video analysis should be used.
For photocell separations that were twice as large,
the speed errors were around half the size. T his is
to be expected, since, for sim ilar tim ing errors, the
speed errors will be inversely proportional to photocell
separation. T hus there is a case for having the separation
as large as is practicable. For m easurem ent of long jum p
or pole vault approach speed, however, the changes in
speed over the last 10 m m ay be required rather than the
average speed over the whole interval. If a single beam
system is used to calculate the average speed over the
last 5 m and the previous 5 m , the speed errors could be
Yeadon et al.
Close to two stride lengths
RM S 1
M ean ± s RM S RM S 2 F -test
- 0.13 ± 0.18 0.22 1.30 P > 0.05
- 0.03 ± 0.15 0.15 1.53 P > 0.05
- 0.01 ± 0.10 0.10 1.75 P < 0.05
0.00 ± 0.07 0.07 0.91 P > 0.05
less than 0.1 m ´ s - or nearly tw ice as large depending on
1
the stride length. E stim ates of the m ean length of each
of the last four strides of the long jum p approach for
12 m ale athletes at the U S C ham pionships are given in
Hay et al. (1986). The sum of the W rst two stride lengths
had a m ean value of 4.69 m (s = 0.17 m ) and the sum of
the last two stride lengths had a m ean value of 4.65 m
(s = 0.19 m ). For a photocell separation of 5 m , the root
m ean square speed error of a single beam system would
be close to 0.1 m ´ s - and even sm aller if the longest
1
break criterion was used. For 12 W nalists in the wom en’ s
long jum p event at the 1984 O lym pic G am es, the m ean
stride length over the last four strides was 2.17 m (Hay
and M iller, 1985). W ith a photocell separation of 5 m ,
this would lead to errors > 0.2 m ´ s - , even with the
1
longest break trigger criterion. In such a case, it would
be necessar y to set the separation to around twice the
expected stride length.
For a double beam system , little diV erence in accur-
acy exists w hether the separation is close to two stride
lengths or not. Additionally, the longest break criterion
will im prove the accuracy of speed estim ates for only
a sm all percentage of trials. In addition to its greater
accuracy, the advantages of a double beam system
include ease of operation and sim plicity of data pro-
cessing.
It should be recognized, however, that the errors in
speed estim ates obtained from all of the photocell
system s used in this study are generally little better than
0.1 m ´ s - and can be considerably larger (Fig. 3). This
1
suggests that speed estim ates obtained from photocell
system s should be reported to the nearest 0.1 m ´ s - .
1
F rom the results of this study, we m ake the follow ing
recom m endations for obtaining accurate running speed
estim ates from photocell tim ing m easurem ents:
· use a double beam system if possible;
· place single beam s at hip height;
· use as large a photocell separation as possible;
· set the photocell separation to a m ultiple of stride
length;
report speeds to the nearest 0.1 m ´ s - .
·
1
M easuring running speed
R eferences
Ae, M ., M iyashita, K., Shibukawa, K., Yokoi, T. and Hashi-
hara, Y. (1985). Body segment contributions during
the support phase while running at diV erent velocities. In
B iom echanics IX -B (edited by D.A. W inter, R.W. Norman,
R.P. Wells, K.C. Hayes and A.E. Patla), pp. 343± 349.
Ch ampaign, IL: Human Kinetics.
Br  ggemann, G.P. and Kunkel, V. (1990). Biomechanics of
luge start technique. In Proceedings of the Sports B io-
m echanics Section of the B ritish Association of Sports Science s
15 (edited by R.M. Bartlett), pp. 79± 90. Leeds: BASS.
Chandler, R.F., Clauser, C.E., M cConville, J.T., Reynolds,
H.M . and Young, J.W. (1975). Investigation of inertial
properties of the hum an body. Technical Report AM RL-
TR-74-137 (DOT HS-801 430). Dayton, OH: Wright-
Patterson Air Force Base.
Clauser, C.E., M cConville, J.T. and Young, J.W. (1969).
Weight, volume and center of mass of segments of the
human body. Technical Report AM RL-69-70 (AD-710-
622). Dayton, OH : W right-Patterson Air Force Base.
Dem pster, W.T. (1955). Space requirem ents of the seated
operator. Technical Report WADC-55-159 (AD-608-463).
Dayton, OH: Wright-Patterson Air Force Base.
Doolittle, T.L. (1971). Errors in linear measurement with
cinematographical analysis. K inesiology Review , 1 , 32± 38.
Dyas, J.W. and Kerwin, D.G. (1995). A photocell based tim ing
system for studying linear kinematics of running. In
Proceedings of the Sports B iom echanics Section of the B r itish
A ssociation of Sport and E xercise Sciences 20 (edited by
J. Watkins), pp. 29± 32. Leeds: BASES.
Hamilton, N. (1993). Changes in sprint stride kinem atics with
age in master’ s athletes. Jour nal of A pplied B iom echanics ,
9 , 15± 26.
Hatze, H. (1980). A m athem atical m odel for the com-
putational determ ination of param eter values of anthro-
pomorphic segm ents. Jour nal of B iom echanics, 13 , 833± 843.
Hay, J.G. and M iller, J.A. (1985). Techniques used in the
transition from approach to takeo V in the long jum p.
Inter nationa l Jour nal of Sport B iom echanics , 1 , 174± 184.
Hay, J.G., M iller, J.A. and Canterna, R.W. (1986). The
techniques of elite male long jumpers. Jour na l of B io-
m ech anics , 19 , 855± 866.
Ikai, M . (1968). Biomechanics of sprint running with respect
257
to the speed curve. In B iom echanics I (edited by J. Warten-
weiler, E. Jokl and M. Hebbelinck), pp. 282± 290. Basle:
Karger.
Jensen, R.K. (1978). Estimation of the biom echanical
properties of three body types using a photogramm etric
m ethod. Jour na l of B iom echanics , 11 , 349± 358.
Karara, H.M . (1980). Non-metric cam eras. In D evelopm ents
in C lose Range P hotogram m etr y , Vol. 1 (edited by K.B.
Atkinson), pp. 63± 80. London: Applied Science Pub-
lishers.
Kerwin, D.G. (1995). Apex/target high resolution video
digitising system. In Proceedings of the Sports B iom echanics
Section of the B ritish Association of Sport and E xercise Science s
20 (edited by J. Watkins), pp. 1± 4. Leeds: BASES.
M ero, A. and Komi, P.V. (1985). E V ects of supram aximal
velocity on biom echanical variables in sprinting. Inter -
nation al Jour na l of Sport B iom echanics , 1 , 240± 252.
Nigg, B.M ., Bahlsen, H.A., Luethi, S.M . and Stokes, S.
(1987). The in X uence of running velocity and m idsole
hardness on external impact forces in heel± toe running.
Jour na l of B iom echanics , 20 , 952± 959.
Poulm edis, P., Rondoyannis, G., M itsou, A. and Tsarouchas,
E. (1988). The in X uence of isokinetic muscle torque
exerted in various speeds on soccer ball velocity. Jour na l of
O rthopaedic and Sports Physical Therapy , 10 , 93± 96.
Richards, J., Farrell, M ., K ent, J. and Kraft, R. (1985). Weight
transfer patterns during the golf swing. R esearch Q uarterly
for E xercise and Sport , 56 , 361± 365.
Viitasalo, J.T. and Korjus, T. (1987). M essung der Abwurf-
geschwindigkeit und des Abwurfwinkel beim Speerwurf.
Leistung ssport , 17 , 39± 41.
Yeadon, M .R. (1990). The simulation of aerial movement.
Part II: A mathematical inertia model of the human body.
Jour na l of B iom echanics , 23 , 67± 74.
Yeadon, M .R. and M orlock, M . (1989). The appropriate use
of regression equations for the determination of segmental
inertia param eters from anthropometric measurements.
Jour na l of B iom echanics , 22 , 683± 689.
Zatsiorsky, V., Seluyanov, V. and C hugunova, L. (1990). In vivo
body segment inertial parameter determination using a
gamm a-scanner method. In B iom echanics of H um an M ove-
m ent: A pplications in Rehabilitation, Sports and Ergonom ics
(edited by N. Berm e and A. C appozzo), pp. 186± 202.
Washington, DC: Bertec.