You are on page 1of 25

2 CHAPTER

Flexibility - r on-line systen underwater, iI Allows visua permanent re On-line syste

VIDEO USING ANALYSIS MOTION


Corl J. Poyfon

Given the advantag able future, an imp Video analysis of in nature. Qualitat observation of the 1 on a TV monitor < and frame-by-fram displayed simultan The purpose of th movement being ol It may also be use that need to be qui

INTRODUCTION
measurement techFor many decades,cinematography was the most popular cine cameras have motion. human of analysis r", those involved in tf,e becauseof their much "iq". traditionally been considered superior to video cameras over the last decade, However, rates. frame hiiher g*"i., pi.,lre resoluti,onand which now make technology in video made considerable advances have beeln are now able to cameras video Modern cine. to video an attractive alternative quite as good as cine) and deliver excellent pl.t.rr. quality (although still not comparable to high-speed high-rp."d -od.is .u.r u.ii.u. frames rates at least involves -no processing recording video most film, .ii. .ur.r.r"s. Unlike cine playback and analysis' immediate for available are time and the recorded images Vid.o,up.,afeveryinexp"ensivewhencomparedtothehighcostofpurchasing in video camera u.rJ p.o..rring of .i.t. fim. fhe significant-improvements.made over the hardware the price in fall substlantial a i..hrrology, clupled with _of virtually redundant in sport and olr, a..ii., h", l.d to cine cameras becoming ixercise biomechanics. usually made by video recordings of sport and exercise activities are of an individual's biomechanist, i., ori.r to ,r.,dertake a detailed analysis an attractive provide (Chapter 3) systems on-line movement patterns' Although motion video motion-data' capturing of aLtthod alternative to video' "s analysis motion has a number of practical advantages over on-line "rrulyri, including: o r cheaper than Low cost - video analysis systemsare generally considerably on-line systems. can be conducted Minimal interference to the performer video analysis attachment of e'8' performer' the to disiurbance any for without the need reflective markers.

Quantitative video recording to approach requires qualitative analysi processing procedu involves manually or more points fot Typical landmark joint centres of rol of foot), or exterr ordinates resulting before being used Additional kinema computing the firr However, the accu the appropriate dt The kinematic inf, performance parat can then be con performers (e.g. fa (e.g. to evaluatetl In order to u technique, more d common approac method involves joint reaction for< through video, or computational pr< angular accelerat

VIDEO 9 USING ANALYSIS MOTION . where some Flexibility - video analysis can be used in environments outdoors, e.g. effectively, operate to unable te on_line systems would underwater, in comPetition. - video cameras provide a Allows visual feediack to the performer permanenr record of the movement that can be viewed immediately. performer' bn-line systems do not generally record the image of the

e
iI

e o

d rd
tq

.s. rg
ta

he nd by
Ll's ive ,on rsis

for the foreseeGiven the advantages listed above, video analysis will remain, and exercise. in sport technique of analysing method able future, an important quantitative or qualitative be may technique oi a person's Vid.o "rr"iysis structured and systematic a detailed, involves in nature. Qualitative analysis is displayed image video The pattern. movement observationtf th. p.rformer's motion slow real-time' in observed and screen on a TV monitor or computer are views' side and front e'g' images, multiple and frame-by-frame. Often' undertaken. to be analysis complete -or. allow to displayedsimultaneously " quality of the Thi purpose of this typ. of analysis is often to establish the performer' to the feedback some provide to order in ,nou.In.rr, being obre.rred parameters performance key the identifying of means It may also be Jsed a, a that need to be quantified and monitored in future analyses' detailed measurements from the euantitativi analysis involves taking to be quantified. This p"."-etefs perform".r.. key video recording to .nubl. ,opl,isticated--hardware and software than for a approach ,.qrrir., -or. and data qoutit"tlu. arr"lysis and it is iital to follow the correct data capture as it often pro..rrirg procedures. Quantitative_analysiscan be time-consuming (typically eighteen inuolu* ir"ro"lly digitising a number of body landmarks images. of number latge a over model) body f,rll or more points fo, u .video represent to assumed are those digitisation for selected Typical landmarks (e.g. knee joint centrei, segmental endpoints (e.g. end ioint...rtr., of rotation cooi foot), or external obj..ts (e.g. a sports implement)' Two-dimensional smoothed and ordinatesresulting frorn the digitising process are then scaled histories' displacement-time angular and inear calculate to beforebeing or.d is obtained by Additional tine-atic information (velocities and accelerations) data. uting the first and second time derivatives of these displacement unless ,u.r,"rn. accuracy of these derivatives will be severelycompromised Chapter 7)' appropriat. data processing techniques are used (discussedin key quantify kinematic information obiained from video can be used to jump)' Such.parameters rmance parameters (e.g. a take-off angle during a elite), within vs. (e.g. novice performers be"tween then be'compared

or monitoredover a period of time ,torr-faiigued), ,*.r, 1..g.faiig.r.d .,rs. the effectsof training over a season)' to evaluate

or exercise In order to understand the underlying causesof a given sport tl: ique, more detailed quantitative analysesare often undertaken' *:* obtained reaction forces and .r.', -oln.n,s) from kinematic information The inverse.dynamics eh uid.o, or some other form of motion analysis. data, i'e' linear and time-derivative second require procedures ,it"tiorr"i require lar accelerations,for the body segmentsbeing analysed' and also

tan ;ted Fof


I

in Chapter 9)' This upprou.h is thai of inversedynamics(discussed (e'g' net inuolu.t computing kinetic information on the performer

PAYTON 1O CARLJ. (e'g' mass and moment of inertia)' The valid body segment inertia data Ju., b. subiect to significant errors unless calculated joint momen-tsurra ror.., the kinematic and inertia data' ;;;; i, t"k.r, to. minimise the error in ;;;; inverse dynamics analysis is not an The interpre,u,ron oi the results of provides an kin"-uti. analysis. Inverse dynamics as straightforward "r"i;;; of a[ the ..,urcre, crossing a ioint, but it does not insight into the n., by "ri.., ;.ntact forces o.1h. torque produced alrow the .o,,'po,",io.r"-;it;". a are there joint' Although the around individual muscles, or muscle groups' 1990), the (e.g. winter, approach number of limitationr-," irr. i".[.re'dy.r"-i.s wiih a much better understanding method can still provide the biomeciranist acting during a sport or exercise of the musculo-skeletal forces and torques the movement patterns could be obtained from an analysis of ;;;;;y;;h"n alone. Picture quality

DERATIONS CONSI PMENT EQUI


undertaking a motion Selectionof the appropriate equipment is important when video motion analysis analysis st,rdy ,rring uid.o. The key components of a system are: . r -- of the movement; Video camera - to capture images to record and store the images fromdevice ,ro."g. Recording and video camera itself the camera. This mJy be an integral part of the (camcorder)or an external unit, e'g' hard-disc; pluyb"ck system - to allow the video images to be viewed for qualitative or quantitative analYsis; from the video co-ordinate digitisei - to allow measurementsto be taken images; p.oJ.srirrg analysis software - to enable the user to quantify selected ".rd parameters of the movement.

. . .

Videocomeros
undertaking a biomewhen selecting a video camera with the intention of important features to the chanical analy"sisof a sport or exercise activity' consider are:
a a a a a a a

A video image is n A full video image , up of the odd-num the even-numbered methods: interlaced technique record on scan camera record frame are identical. format. I7ith progrr displaying individua up a video image d extent, determinest There are a nu can sometimes lead camera purchasedir player. The phase a (exceptFrance), Aus Couleur Avec M6mr European countries of pixels. This is r, Standards Committt Japan, and has 525 is therefore essentia that the vertical res than these figures,d, Picture quality is als refers to the number In the past cou on the domestic mar format allows high back on DV tape.H affordable prices an resolution of either is important to chec 7201/1080i or progr for example, the disl described, there are have varying resolut

picture quality frame rate (samPling frequencY) shutter manual high-sPeed manual aPerture adiustment light sensitivity gen-lock capabilitY iecording medium (e.g.tape, hard drive)'

a a a

VHS, VHS-Ca lines. S-VHS,S-VHS Digital 8 and n High Definition either 1280or

ANALYSIS MOTION USINGVIDEOI1 Picture quality A video image is made up of a two dimensional array of dots called pixels. A full video image or frame consists of two halves or fields. One field is made up of the odd-numbered horizontal lines of pixels, the other is made up of the even-numbered lines. Video cameras capture an image using one of two methods: interlaced scan or progressive scan. Cameras that use the interlace techniquerecord one field first, followed by the second,and so on. A progressive scan camera records a complete frame and the two fields that comprise this frame are identical. Some cameras have the facility to capture images in either 'With progressive scan, the option to analyse a movement at 50H2, by format. displaying individual video fields, is lost. The number and size of pixels making up a video image determine the resolution of the picture and this, to a large extent, determines the picture quality. There are a number of different world standards for video equipment; this can sometimes lead to problems of compatibility. For example, a digital video camerapurchased in the USA, may not be compatible with a UK sourced DV 'S7estern Europe player. The phase alternating line (PAL) standard is used in India and China. Sequential (except France), Australia and much of East Africa, Couleur Avec M6moire (SECAM) is the standard found in France and Eastern European countries. Both PAL and SECAM video have 625 horizontal lines of pixels. This is referred to as the vertical resolution. National Television StandardsCommittee (NTSC) is the standard adopted in North America and Japan,and has 525 lines. The maximum vertical resolution of a video image is therefore essentially limited by the video standard used. It should be noted lthat the vertical resolution of a displayed image might be considerably lower thanthesefigures, depending on the specification of the video equipment used. quality is also influenced by the horizontal resolution of the video. This to the number of pixels per horizontal line. In the past couple of years a new video format called HDV has emerged the domestic market and is likely to supercedeexisting standards. The HDV t allows high definition (HD) video images to be recorded and played on DV tape. HDV video cameras are now commercially available at very ble prices and the images produced by these cameras have a vertical ion of either 720 or 1080 lines. When purchasing an HDV camera, it important to check what mode(s) it can record and playback in (interlaced: '$Tithin each of the world video standards just example,the display device. ;ribed, there are a number of video recording formats available and these

tf
'e o )d

to ensurethat it is compatiblewith, 720p11.080p) 1080ior progressive:

teto

varyingresolutions: VHS,VHS-C and 8mm formats eachdeliveraround 240-:260horizontal lines. S-VHS, S-VHS-Cand Hi-B video provide around 400 horizontal lines. Digital8 and miniDV deliverat least500 horizontal lines. HighDefinition(HD) videogiveseither720 or 1080horizontallines(with pixels per line). either1280 or 1.920

J. PAYTON 12 CARL

(a)

(b)

rotes up to of frome (Pholron APX) copoble Ultimo Fostcom videocomero 2.1 (o) High-speed Figure (b)Comero (1024x 1024 pixels); Processor Unit 2OOOHz ot fullresolution Some specialist video cameras (e.g. Photron FastcamUltima APX in Figure 2.1) can record images with resolutions higher than those describedabove. It should be noted that even within a given recording format, e.g. miniDV, the quality of the video image can vary considerably. The resolution of the camera is largely influenced by the quality of its image sensor - the component that converts the light from the object into an electrical signal. The most common type of image sensor is the charge-coupled device (CCD). Most domestic video cameras have a single CCD chip, but some higher quality models have three CCDs (one for each of the primary colours), which result in an improved picture quality. An alternative to the CCD is the complimentary metal oxide semiconductor (CMOS) image sensor. This sensor requires far less power than a CCD and is now used in some standard and high-speed video cameras. The specification of the camera lens is an important factor in determining picture quality. Digital video cameras will have both an optical zoom range, i.g.20x and a digital zoom range' e.g.400x. It is important to note that once a camera is zoomed in beyond the range of its optical system, the picture quality will drastically reduce and will be unsuitable for quantitative analysis. A...rro.y telephoto lensescan be used to increasethe optical zoom of a digital video camera and avoid this problem. They also allows the user to increasethe camefa-to-subject distance, whilst maintaining the desired image size. This will reduce the perspective error although it should be noted that the addition of a telephoto lens will reduce the amount of light reaching the camera's image sensor. It is important to check how well a telephoto lens performs at the limits of the optical zoom, as this is where image distortion will be most pronounced. \X/ide-anglelensescan be fitted to video camerasto increasethe field of view for a given camera-subject distance. However, such lensestend to produce considerable image distortion and have limited applications in quantitative analyses.

to the number of fu as the sampling freq a frame rate of 251 If the camera captu will be comprised o1 with a vertical resol one field comprised appropriate hardwar separately and seque a second increments For some sport cameras will be too I speed video camera or digital (see Greav frame rates beyond 2 100-500H2 are gen applications. Althoul (e.g. Peak Performan to RAM (e.g. Photrc a computer hard dr camera). One of the maj is the limited recordi with a storage capa at 2000 Hz, provide seconds.

High speed shutter

Frome rote (sompling frequency) 'frame' refefs to a complete image captured at an In video captufe, the term instant in time (Greaves, L995). Thus the frame rate of a video camera refers

For most biomechan speedshutter is esse the amount of time tl light. Modern video c or deactivating the in sampled. \7hen recor is exposed to the lig period of time; this r The extent of the blt analysed. It is important This allows the use as it represents the activity that is being Collection Procedur offer shutter speed noted that not all v models that incorpo

ANALYSIS MOTION USING VIDEO ] 3 to the number of full images it captures per second (this is often referred to as the sampling frequency of the camera). Standard PAL video cameras have a frame rate of 25H2, whereas NTSC cameras have a frame rate of 30H2. If the camera captures using the interlaced scan method, each video frame will be comprised of two video fields (an A and B field). For a video image with a vertical resolution of 480 lines, each field would consist of 240 lines, one field comprised of the odd lines, the other of the even lines. With the appropriate hardware or sofrware, it is possible to display the video fields separatelyand sequentially thus enabling measurementsto be taken at 1/50 of a second increments (or 1,/60of a second for NTSC), but at reduced resolution. For some sport and exerciseactivities, the frame rate of conventional video cameraswill be too low and a high-speedvideo camera may be required. Highspeed video cameras, as with conventional video cameras, can be analogue or digital (see Greaves,1995 for more detail). Although video cameras with frame rates beyond 2000H2 are commercially available, cameras with rates of 100-500 Hz are generally adequate for most sport and exercise biomechanics applications. Although some early high-speed video cameras recorded to tape (e.g.Peak Performance HSC 200 PS),most models now either record the images to RAM (e.g. Photron Fastcam Ultima APX shown in Figure 2.1) or direct to a computer hard drive via a Firewire (IEEE) port (e.9. Basler 602f 100H2 camera). One of the major limitations of high-speed cameras that record to RAM , is the limited recording time available. For example, a high-speedvideo camera with a storage capacity of B Gb, recording with a resolution of 1,024 x 1024 at 2000 Hz, provides a maximum recording duration of approximately three

seconds.
r q 'b g,

Highspeed shutter For most biomechanical applications,a video cameraequippedwith a highof a camera that controls The shutteris the component speed shutteris essential.
amount of time the camera's image sensor (e.g. CCD, CMOS) is exposed to t. Modern video cameras use electronic shuttering, which involves activating deactivatingthe image sensorfor a specifiedtime period, as each video field is . lfhen recording movement using a low shutter speed,the image sensor exposedto the light passing through the camera lens for a relatively long iod of time; this can result in a blurred or streaked image being recorded.

at re is. :al he ,ill of rge Lit's ed. for rid;es.

'shutter speed' (this term is a misnomer is allows the user to select a it represents the time the shutter is open) that is appropriate for the ivity that is being analysed, and the prevalent lighting conditions (seeData ion Procedures section of this chapter). Typically, a video camera will

extentof the blurring would dependon the speedof the movementbeing lysed. It is important that a video camerahas a manual shutter speedoption.

shutter speedsranging from tl60-1,/4000 of a second.It should be that not all video cameras offer a manual shutter function. Camera
that incorporate a Sports Mode function should be avoided because

PAYTON 14 CARLJ. for fast-moving the shutter speed associatedwith this is often inadequate activities. Recording medium

Monuol iris and low-light sensitivity the aperture The iris is the element of the camera's lens system that controls falling on light (the adjustable gap in the iris) in order to regulate the amount of (large lens iir. i-is. r.rrrJr.'If too mucL light is permitted to pass throughthe light little If too image. orrerexposed for too long, the result-will be "n Video "p.r,"rJl, (small aperture), the image will be underexposed. f'urr", through th. l"i, automatic aperture control that continually adjusts to have genlrally i"-.ru, have a manual ensure the image is correctly exposed. Some camera models This is sometimes override that aliows the user to specify the aperture setting. when a high .r...rr"ry when conducting biomechanical analyses.For example' would aperture iris shutter speed setting i, .r.!d.d in low light conditions' the drawback The fr"u. ,o b. op.n.d i"id". thun it would be in automatic mode. results in a more oi aoi"g this is the increased noise level in the image, which 'grainy' picture. in order video cameras each have a minimum light level that they require a minimum with to produce an image. This level is expressedin lux. A camera than one conditions light low in better perform will 1 lux ,r"1,-r..-of illumination witha3luxrating.

Images from video ce tape, for example, S recording formats ha DVD are more gear undertake a quantita recording camerasa a hard disc drive (H

ond sk Recording

Genlock copobility activation of the For three-dimensional video analysis, it is desirable for the synchronised, that is, shutters of the two (or more) cameras to be perfectly linking the cameras for the cameras to be gen-locked. This involves physically do not gen-lock cable."Unfortunately, most standard video camcorders *irt do models " expensive fr"r. ifr". facility to be gen-locked, although some more cameras video If offer this feature 1e.g. banon XL H1 HDV 1080i camera). from each obtained co-ordinates two-dimensional th. gen-lock.i, cannot be data and then the of the camera views must be synchronised by interpolating time lag The shifting one data set by the time lag between the camera shutters. (e.g. at camera of the wifl bJ no more than half the reciprocal of frame rate the determining )sn ,the time lag will be <20-r;. th. simplest method of this srhere cameras' time lag is to have a timing device in the field of view of all a.method involving competition, at a filming when for .*u-pi. i, .rot iorrlble, analysis of th. .o-ordinates of all the digitised body landmarks, -",h.-"rical " certain commercial nu, u..r, proposed (ieadon and King, 1,ggg).Alternatively, analysis software-packages,for example SIMI"Motion, will ulJ.o .up,rrr. ".rd if the video automatically measure the time lag b.twee.t camera shutters' via the drive' hard i-ug., from the cameras are simultaneously captured _to a the shift phase ,ofti""r., in real-time. The software will also interpolate and be to reconstruction two-dimensional co-ordinates to enable three-dimensional undertaken.

A video camera that of options, depend qualitative analysis videotape in real-tin video playback sys software. Alternativ the tape to a comp video file (e.g. AVI, of appropriate sofn achievable when plz multiple video clips be undertaken, if a installed. Video images RAM or Flash Mer hard drive, where tl The processof captt time or at some poil is taken will be det of the camera and 1 tape, or which hav can be done post-r the only option, as the system. In most speed cameras),cap appropriate sofftva or mofe camerasca video imagesto a co

Specificotion of com

For real-time video IEEE-1.394 connect is not an integral p

VIDEO 15 USING ANALYSIS MOTION Recording medium Imagesfrom video cameras have traditionally been recorded onto some form of t"p., fot example, S-VHS and miniDV. In recent years a number of alternative reiording fot-ats have emerged. Video cameras that record straight to a small DVD are more geared toward the home movie-maker, than those wanting to undertake a quantitative analysis of movement. More viable alternatives to tape are those with built-in memory. This may be in the form of recording."-.r", a hard disc drive (HDD), internal memory (D-RAM) or Flash Memory'

:e In

ht ]o to ral res gh

ond storogedevice Recording


A video camera that records the images to tape provides the user with a number of options, depending on what type of analysis they are performing. For a qualitative analysis, the recorded movement can be viewed directly from the videotape in real-time, slow motion or as a still image, using an appropriate video flayback system, without the need for any computer hardware or softwaie. Alternatively, the user may choose to capture the video images from the tape to a computer hard drive, where they are stored in the form of a video hle (e.g. AVI, MPEG, etc.). This is an attractive option as, with the aid of appropriate software, images can be presented in ways that are not easily achievabL when playing back directly from tape, for example, the display of multiple video clips simultaneously. It also enables a quantitative analysis to be undertaken, if appropriate digitising, processing and analysis software is installed. video images that are recorded to a camera's hard disc drive (HDD), RAM or Flash Memory are usually transferred subsequently to a computer hard drive, where they can be displayed or processedfor quantitative analysis. The processof capturing video images to a computer can either be done in realtime or at some point following the filming session.Which of these approaches

ild ,ck )re der um


0ne

the t is,
ICTAS

r not ls do oeras each , then ne lag ).g. at rg the re this olving narks, nercial n, will video via the rift the ntobe

by a number of factors includingthe specification is takenwill be determined


the camera and the filming environment. For video cameras that record to or which have their own hard drive or memory, capture to computer be done post-recording. I(ith the majority of high-speed cameras this is only option, as the required data transfer rate exceeds the capability of ryri.-. In most situations with standard 50Hz cameras (and some higher ,d cameras),capture of video to computer can be done in real-time. With from two late softwaie, and the requisite connectivity, video sequences ror. .u-.rus can be captured simultaneously in real-time. When capturing .eo imagesto a computer, the following practical issuesneed to be considered:

of computer a Firewire real-time video capture from standard digital video cameras. If this i-Link). 1394 connection is required (often referred to as DVin or via connected not an integral part of the computer, a Firewrre hub can be

r
PAYTON 16 CARLJ. t h e U S B o r P C l p o r t . A l t e r n a t i v e l y ' a P C M C I A F i r e w i r e c a r 2'0 d c aot n bEthernet eused(for aigitat camera models' a USB laptops). For some h;;;;;; data' For capture from an,analogue source' to do#nto"d pJr, i-r.n-lr.d "ldtl is needed.This *ust be able to capture in a some form of video ."pi"r. u."rd digitising software. resolut'J" inu, is compatible with the file format "nd n."ut sufficiently fast processor and adequate All modern."t";;;;;u " RAMtomanagethedatatransferratesinvolvedinstandardd igitalvideo rates up to 400 Mbitis which is more than capture (Firewire ,,lppo"' transfer consideration is the available hard adequate for DV viJeo). An important. five second video file captured at a disc space. The size ""..r"pressed "ii" i' io.rt 1gMb. Four minures of uncompressed resolution of 720" szrli".r, hard disc space. will therefoie require almost 1 Gb of ;;;;;ffi;

For quantitatil computer. Here, the card and displayedc by the specification< settings, monitor res

digit Co-ordinote

Copture sofrwore C a p t u r e s o f t w a r e i s u s e d t o c o n v e r t o r e n c o d e t h e v i d e o t o t h e are r e q critical. uiredfile .codec,. The capture Settlngs used within the software format or (e.g. AVI' lo1-1i of. the captured video file For quantitur,u. should user The ",.,"i|r;';;;with the digirising software. MPEG) must be .#p",iui. s h o u ld not q u a l i t v q . u a l i t va " v a i l a b i eI'm a g e ;;;;;il".rp,ur..in'tf,t-higt"" formats' size, by using high-compression be compromlsed ror the saf,e of fiie unless absolutelY necessary'

Video ploybocksystem
'flicker Avideoplaybacksystemisrequiredto.displaythevideoi-:q:tforqua litative should be capable of displaying irr. br,.." or quantiratiu. to be played in slow ".ruryJJ. video sequences free' still images. f, ,nt"fa "fst "llo* motion and in real-time' an analogue or digital video player-recorder For qualitatiu. """ly'i', ui"Lle option. This should be equipped i, rV- -iriio. (VCR) linked ,o " For " pause and picture advance-functions' with a jog-shuttle dial to control analoguevideo,ro.nu'S-VHS,afour-headVCnisnecessary forastablestill will enable individual video fields to image. Some profes,i"""f gt"Jt VCRs with 50 images per second (compared be displayed, ,no, p,luilitig 'ht user

To undertake a deta This device enables the video image, fo based co-ordinate d the still video imag that is manually cc when selecting a vi refers to the minim system is able to dt to which the co-or systems offer consid in early systems.Tl pixel cursor. For ex Ltd, Coventry, UK resolution of 720 (h linearly using the s< 3 x magnification, 1 0.05%. The TARC University combine a digitising resoluti that, unless the reso 'pixilated' at h very

DATA COLLE
'When

is important f1:11]ll *,rft lT p.. ,..o,td-o" -o'Jdornt'tic videollalers)' Jhit quality on a video Tht picture ^;""ri,y very slow

when analysingall but -movemtn"' of .r the sourcetape, the specification monitor i, irrfl.r.r..jl;;ir. device playback the int typt'of video cableusedto link video playbackdevice, (RGB)' itatt of quality: composite'S-video'Scart to the monito, f ,n "r.i,tJilg itself' .o-fon.",, DVI' HDMI), and the monitor Traditionalcnr-""i,.rsgenerallyofferexcellentpictureresolutionbut LCD Monitors vary in their resolution cannot directly dirpl"y ;;i;itul i,r.... 1'024x 768; HD monitor: (e.g.VGA *ont,or, 640 x"480; XGA monitor: 1'366x768).SomemodelsofLCDmonitorscanonlydisplayanalogueSources' jlgi,"t sources' and somecan displayboth' -*. ""fy

conducting r followed carefully, minimise the systen when undertaking procedures are still record of the perfor Quantitative v The former approz being analysed is motion. Any meas subject to perspect appear to be two-<

USINGVIDEO ANALYSIS MOTION

'17

for 'net rce, na rate ,deo :han rard ata ssed

analysis, video playback will be via a laptop or desktop For quantitative ^Here, the video data are processedthrough the computer's graphics computer. ."ri"rrd displayed on the monitor. The quality of the image will be influenced by the specification of the graphics card, the video playback codec' compression settings, monitor resolution' and the digitising software'

digitiser Co-ordinote
To undertake a detailed quantitative analysis,a co-ordinate digitiser is required. This device enables two-dimensional (x, y) co-ordinates of specified points on the video image, for example, anatomical landmarks' to be recorded. Videobased co-ordiiate digitis.ri ut. essentially software applications that display the still video image on a computer screen and overlay this with a cursor that is manually controlled by the user' The most important consideration when selecting a video digitising system is its measurement resolution. This refers to the minimum separation between two points on the screen that the system is able to detect. The digitiser resolution affects the level of precision to which the co-ordinates can be measured. Current video-based digitising systemsoffer considerably higher measurement resolutions than were available in early systems.This is achieved through a combination of zoom and a subpixel cursor. For example, QUINTIC Biomechanics 9.03 (Quintic Consultancy Ltd, Coventry, UK) displays the non-magnified standard video image at a resolution of 720 (horizontal) by 526 (vertical). This resolution can be increased linearly using the software's zoom function (up to a maximum of 10x). At a 3x magnification, this provides a measufement resolution of approximately 0.05%. The TARGET video digitising system developed at Loughborough University combines a 4x magnification with a sub-pixel cursof to produce a digitising resolution of 12,288 x9,216 (Kerwin,1.995).It should be noted that, unless the resolution of the captured video is high, the image will become very'pixilated' at high magnifications.

d file ,tica1. AVI, rould I not mats,

itative flicker r slow ,corder uipped rs. For blestill eldsto mpared portant a video rtion of < device (RGB), tion but solution nonitor: sourcest

PROCEDURES COLLECTION DATA


Vhen conducting a quantitative video analysis, certain procedures must be followed carefully, at both the video recording and digitising stages' to minimise the systematic and random errors in the digitised co-ordinates' Even ,when undertaking a qualitative video analysis, many of the video recording proceduresare still peitinent as they will help to obtain a high quality video recordof the performance' or three-dimensional. Quantitative video analysismay be two-dimensional The former approach is much simpler, but it assumes that the movement - the plane of being analysei- is confi.red to a single, pre-defined plane motion. Any measurements taken of movements outside this plane will be rbject to perspective error, thus reducing their accuracy. Even activities that ppear to be two-dimensional, such as a walking gait, are likely to involve

I8

J. PAYTON CARL

movements in more than one plane; a two-dimensional analysis would not enable these to be quantified accurately. Three-dimensional analysis enables the true spatial movements of the performer to be quantified. This approach eliminates perspective error, but the video filming and analysis procedures are more complicated, and the equipment requirements are also greater.

videorecording Two-dimensionol
The following guidelines are designed to minimise the systematic and random errors present in two-dimensional co-ordinates, resulting from the video recording stage. This will increasethe accuracy of any parameters subsequently obtained from these co-ordinates. The guidelines are based on those previously reported in Bartlett, 1.997b, and in earlier texts (Miller and Nelson, 1'973; Smith. 1975\.

Equipment set-up Mount the comera on o stoble tripod and avoid ponning The standard approach in a two-dimensional analysis is for the camera to remain stationary as the performer moves through the field of view. This enables the movement of the performer to be determined easily relative to an external frame of reference. Two-dimensional filming techniques involving panning or tracking cameras have been used when the performance occurs over a long path (for example As thesemethodsinvolve the cameramoving Gervaiset aI.,1989; Chow, 1,993). (rotating or translating) relative to the external frame of reference,mathematical corrections have to be made for this movement if accurate two-dimensional co-ordinates are to be obtained. Maxi mi se the comero-to'subiect distance The camera must be positioned as far as is practically possible from the performer. This will reduce the perspective error that results from movement outside the plane of performance (seeFigure 2.2). A telephoto zoom lens will enable the camera-to-subject distance to be increased whilst maintaining the desired image size. Note that image quality will be reduced if a digital video camera is positioned beyond the limit of its optical zoom system. Moximise the image size To increase the accuracy during digitising, the image of the performer must be as large as possible. Image size is inversely proportional to the field of view of the camera. The camera should therefore only be zoomed out sufficiently for the field of view to encompass the performance path, plus a small margin for error. For events that occur over long performance paths, e.g. triple ;'ump, a single stationary camera would not provide an image size suitable for

(a)

Figure2.2 Apporentdir to-subiect distonce of 3 n width oport

quantitative analys cameras, or a pann.

Focus the camero m Most video camer overridden. In most For a well-focused manually focus, an<

Align the opticaloxi Any movements thz not be subjectto pe parallel to the came As no human move of the activity is of 1 can then be positio lens to the geomet of the optical axis orthogonal to the p triangles (trianglesv optical axis is orthc have a detrimental r 2003). Even with a outside the plane o Figure 2.3.

MOTION ANALYSIS USING VIDEO 19


I
t

n o
l

v
i;

(a)

(b)

lo
CS

Figure 2.2 Apporentdiscreponcy in the lengths of wo identicalrodswhen recordedusingo comeroio-subjectdistonce of 3m (imogeo) ond 20m (imogeb). Note thotthe rodsore being held shoulder widthooort

al nrle ng :al ral

quantitative analysis. In such situations, the use of multiple synchronised cameras,or a panning/tracking camera method, would be required. Focusthe comera monually Most video cameras have an automatic focus system that can be manually overridden. In most situations, the camera should be set in manual focus mode. For a well-focused image, zoom in fully on an object in the plane of motion, manually focus, and zoom out to the required field of view. Align the opticol oxis of the comero perpendicular to the plone of motion Any movements that are performed within a pre-defined plane of motion will not be subject to perspective error at the digitising stage,provided this plane is parallel to the camera image sensor (perpendicular to the camera optical axis). As no human movement is truly planar, it is essentialto establish which aspect of the activity is of primary interest and in which plane this occurs. The camera can then be positioned accordingly. Marking a straight line from the camera lens to the geometric centre of the field of view can represent the direction of the optical axis. Various methods can be used to align the optical axis orthogonal to the plane of motion. A common approach is to use right angle triangles(triangles whose sides are in the ratio 3:4:5). Failure to ensure that the optical axis is orthogonal to the plane of motion, even by a few degrees,can have a detrimental effect on the accuracy of the analysis (Brewin and Kerwin, 2003). Even with a correctly aligned camera) movement will inevitably occur outside the plane of motion. The effect on measured angles is illustrated in Figure2.3.

:he
lnt

be lity its

:be , of. for for


ilPt

for

20

PAYTON CARLJ.

tilted more than a Kerwin, 2003).

(a)

(b)

Thetruevolueof outside theploneof motion. when movement occurs of ongles Figure2.3 Distortion thon90" (A') ond ongleB to be greoter o). In imogeb, ongleA oppeors A ond B is 90' (imoge ongles thon 90" (B'),os the fromeis no longerin the ploneof motion to be less oppeors

Se/ecf on approprio In activities such a distal body segmen speed should be se the fastest moving t speed depends on r such as a grande p a second should be or a swimming sta appropriate; for fas speedof 1/1000 of An increasein in image quality, for obtain the best poss must be provided su excessively.

Record o vertical reference To enable a true vertical (and horizontal) frame of referenceto be established at the digitising stage, a clear vertical reference, such as a plumb line, must be recorded after the camera set-up has been completed. Any good video digitising system will correct for a non-vertically aligned camera, using the co-ordinates of the vertical reference. Record a scoling obiect An object whose dimensions are accurately known must be recorded in the plane of motion. This is to enable image co-ordinates to be transformed to object-space (real world) co-ordinates following the digitising stage. Recording of the scaling object(s) must be done only after the camera set-up is complete. The use of both horizontal and vertical scaling objects is essential,because the computer may display the image with an aspect ratio (ratio of the width to the height) that distorts it in one dimension. To minimise the error in the scaling process, the dimensions of the scaling objects should be such that they occupy a good proportion of the width and height of the field of view. For a given digitising error, the scaling error will be inversely proportional to the length of the scaling object. For field widths greater than 2-3 m, scaling is usually done using the known distance between two or more reference markers or control points, positioned in the plane of motion. In some circumstances it is not possible to align the camera optical axis correctly with the plane of motion, for example when filming in a competition. Here, digitisation of a grid of control points, placed in the plane of motion, can be used to correct for the camera misalignment. This method is called 2D-DLT and has been shown to provide significantly more accurate reconstruction of two-dimensional co-ordinate data than the more commonly used scaling techniques, particularly when the optical axis of the camera is

Ensurecorrect lightin If filming indoors, fl, level. Bartlett, Igg7z the plane of perforn should provide adeq otten preferable to I inevitably less predic sun will restrict wher a good contrast with rVhen filming indoor preferred. Video cam light sources (e.g. da white balance, which

Select on oppropriate Standard PAL video c effectively be double Most high-speedcam depend on the frequ dependent variables b detail) states that the that of the highest fre rate should be much I higher). A sufficiently hir and minimum displac Key events rn a pertor swing) are recorded.A

MOTIONANALYSIS USING VIDEO 21

tilted more than :r few degreesrelative to the pl:rne of motion (Brer,vinand I(erwin, 2003). Se/ecf on oppropriote shutter speed ond operture In activities such :rs running, jumping, throwing and kicking, it is the rnost distai body segments,the hands and feet, which move the quickest. A shutter speed should be selectedthat is sufficient to provide a non-blurred image of (or sports implements).Thc choice of shutter the fastestrnovirrgbody segments speed depends on the type of activity being recorded. For slow movernenrs, such as a gr:rnde p1i6 in b:rllet or walking, shutter speedsof 71750-11250 of a second should be :rdequate;for moderately fast activities, such as rr-rnning or a swimming start, shutter speeds of 11350_11750 of a second are more appropriate; for fast erctivities such as a golf swing or a tennis serve,a shr-rtter speedof 1/1000 of a secondor: above rnay be needed. An increarse in shutter speed will :rlways be accompaniedby a decrease in imarge quality, for given tighting conditions and cirmera aperture setting.To obtain the best possibleimagesat the required shr-rtter speecl, sufficientlighting must be proviclcdsuch that the camera iris aperture does not have to be opened excessively. Ensurecorrect lighting of the performer If filming indoors, floocllightsare often neededto achievethe required lighting level. Bartlett, 1997a, sugplests that one floodlight positioned perpendicularto the plane of performance, :rnd one to eirch side :rt :rrouncl 30' to the plane, should provide ildequate illumination. Filming outdoors in natural daylight is often prcferable to filming uncler ar:tificialiights, but natr-rrallight levels are inevitably less predictable.'Jfhcn filming in direcr sunlight, rhe position of the sun will restrict where the cirmeracan bc located.The background musr provide a good contrast lvith the performer and be as plain and unclutterecl as possible. \ff/henfilming indoors with floodlighting, a dark, non-reflectivebackground is preferred.Vicleo cirrner:rs often havc :.rmanually :rdjust:rblc setting for clifferent (e.g. daylight, fluorescentlirmps, soclium or mercury lemps) .rnd light sor,rrces white balance,which cirn be uscd to enhancethe coiour rendition. Selecton oppropriote frqme rqte Standarcl l'}AL video camerash:rveir fired frame retc of 25H2,,although this can effectively bc cloubled,proviclcclthe c:rmera uses thc interlaced scan method. Most high-spccdcirmerash:rvc adjustirbleframe rates.The fr:rme rate used will depend on the frequency content of the moverncnt being an:rlysed,ancl thc clepenclcnt variablesbeing studied. SamplingTheorum (seeChapter 7 for more (frarne rate) rnust be at least double detail) statesthartthe sarnpling frecluerrcy that of the highest frequency prcsenrin the activity itsclf. In re:rlity, the fr,rmc r:rtc should be n'rr-rch higher than this (Challis et al., 1997) suggesr8-10 tin-res hieher). A sufficicntly high frarne rate will ensurethat the instancesof maximum (linear ar.rd irnd minimnm displ:rcement angr-rlar) of a joint or lirnb, and of other key events in a performancc (e.g. heel-strikein mnning, birll impact in :r golf s r ,irr r g ). r r t 'r t ' c o r d e d . A r r i n c r e r s ei n the frame rate will also serveto improve the

of IB

rd
DC

1il
CS

he to
n' D o -

lse to

ng rpy ren of )ne rol cal


la

lne rocl 'ate

,nly
lis

22

CARLJ. PAYTON

precision, and therefore the accuracy, of temporal measurements,for example, the phase durations of a movement. This is particularly important where the phases are of short duration, for example, the hitting phase of a tennis serve. Some suggestedframe rates for a variety of activities are given below:
a a a a

25-50 Hz - walking, swimming,stair climbing. 50-100 Hz - running, shot put, high jump. - sprinting,javelin throwing, football kick. 1.00-200F{2 golf swing,parry in fencing. 200-500 Hz - tennisserve,

It should be noted that these frame rates are only offered as a guide. For a given activity, the appropriate frame rate should be determined by the frequency content of the activity and the dependent variables being measured. For example, a quantitative analysis of the interaction between the player's foot and the ball during a football kick would require a frame rate above 1000 Hz, whereas a rate of 25 Hz would be more than adequate for determining the length of the final stride during the approach to the ball. The effect of using different frame rates on the recording of a football kick is shown in Figure 2.4.

Participant preparation ond recording triols The health and safety of the participant is paramount during any testing. Informed consent should always be obtained from the participant (see BASES Code of Conduct in Appendix 1) and completion of a health questionnaire is often required. Sufficient time must be allocated for a warm-up and for the participant to become fully familiar with the testing environment and testing conditions. The clothing worn by the participant should allow the limbs and body landmarks relevant to the analysis to be seen clearly. The careful placement of small markers on the skin can help the analyst to locate body landmarks during digitising, but the positioning of these markers must be considered carefully. Movement of soft tissue means that surface markers can only ever provide a guide to the structures of the underlying skeleton. Markers are often used to help identify the location of a joint's instantaneous centre of rotation. \Thilst a single marker can adequately represent the axis of a simple hinge joint, more complex joints may require more complex marker systems (see Chapter 3 for more detail on marker systems). The number of trials recorded will depend on the purpose of the analysis and the skill level of the participants. As the movement patterns of skilled performers are likely to be significantly more consistent than those of novice performers ('$Tilliams and Ericsson,2005), they may be required to perform fewer trials in order to demonstrate a typical performance. During the filming it is often useful to record a board in the field of view, showing information such as the date, performer, trial number and condition, and camera settings.

Figure 2.4 Theeffectof c (middlerow) the foot remo

IC

t.
s te
t^

rg 0f in

Figure2.4 The effecto[ comerofrome rote on the recordingof o footboll kick. At 50 Hz (top row] the foot is only seenin contc (middlerow! the foot remoinsin contoctfor four imoges;ot 1000 Hz (bottomrow) the foot is in contoctfor sixteenimoges(not oll
to 'b'

@@@reM

E,S is
he ng dy of .ng llv.
?a to ita 0re for

rsis
led rice rrm .ing narera

24

PAYTON CARLJ.

video recording Three-dimensionol


Many of the procedures described in the previous section for two-dimensional video analysis will also apply when using a three-dimensional approach (selecting an appropriate frame rate, shutter speed and aperture; ensuring correct iightitrg of th. performer; maximising the image size and focusing the camera manually). This section will discuss the main issues that must be considered at the recording stage of a three-dimensional analysis'

set-up Equipment
The essential requirement is to have two or mofe cameras simultaneously recording the peiformance, each from a different perspective. The choice of algorithm or.J to reconstruct the three-dimensional, real world co-ordinates from the two-dimensional image co-ordinates is important as some algorithms place severerestrictions on camera locations. Some three-dimensional reconstruction algorithms rely on very precise positioning of the cameras relative to one another. For example, the method proposed Ly Martin and Pongrantz, 1974, requires the optical axes of the to be orthogonally aligned and intersecting. Such methods can involve i"-..", considerable set-up time and may be impractical to use in some environments, in sports competitions, as they are too restrictive. e.g. " The most widely used three-dimensional reconstruction algorithm used in sport and exercise biomechanics is the Direct Linear Transformation (DLT) aigorithm. This approach does not require careful camera alignment and th-us allows more flexibility in the choice of camera locations' The DLT method determines a linear relationship between the two-dimensional image co-ordinates of, for example, a body landmark, and the three-dimensional, real world co-ordinares of that landmark. A detailed theoretical background to the DLT algorithm can be found elsewhere (e.g. Abdel-Aziz and Katata,"l.971; Miller, Shapiro and Mclaughlin, 1980). To eitablish the relationship between the two-dimensional image co-ordinates and the three-dimensional real world co-ordinates, an object space or performance volume must be defined using a set of control points whose real *orld, three-dimensional co-ordinates are known. This is usually achieved using a rigid calibration frame of known dimensions incorporating a set of visible such as small spheres (seeFigures 2.5 and 2.6). Alternatively, a series -".k.., of discrete calibration poles can be used, provided their real world co-ordinates have been accurately established using, for example, surveying techniques. A minimum of six non-coplanar control points is required for the reconstruction of three-dimensional co-ordinates, but 15-20 control points or more is recommended. The control point co-ordinates must be known relative to three orthogonal, intersecting axes' which define a global co-ordinate system or inertial refei.nce system. This referencesystem is fixed in space and ail three-dimensional co-ordinates are derived relative to this. Images of the control points are recorded by each of the cameras being used in the set-up. These are then digitised to produce a set of two-dimensional co-ordinates for

Figure2.5 Colibrotion I Technologies Inc.)

Figure 2.6 Colibrotio f

each control point compute the 11Dl and position of ea object spaceco-ord from camera 1,,the and 2.2.

xr * CrXr* Yr * CsXr*

For the minimum o camera view. As the

VIDEO 25 USING ANALYSIS MOTION

ral ch ng
n ' ^g D

be

rsly rof
ltes tms

points (Peok Performonce frome(l .60 m x 1.91 m x 2.23 m) with 24 control 2.5 Colibrotion Figure Inc.) Technologies

cise hod the olve lnts,

:din )LT)
and

DLT
nage , real o the

t97t;
(courtesy Sonders) of Ross points frome(1.Om x I .5 m x 4.5 m)with92 control 2.6 Colibrotion Figure

mage space e real using risible series inates )s. )r the points (nown 'dinate ce and of the set-up. ,tes for

eachcontrol point from each camera view. These co-ordinates are used to compute the 11DLT parameters (C1-C11), which relate to the orientation and oosition of each of th. cameras. For control point #1 with real world' objectspaceco-ordinates(Xr, Yr,z1) and with digitised co-ordinates(x1, y1) from camera 1, the DLT equations for that camera are given by equations2.1 and2.2. xr * CtXt -f CzYt -l CtZt't Q (2.1) y 1 * C . s X i - l C 6 Y 1 ' t C z Z t * C g + C s y t X r - F C t o y t Y t I C 1 : y 1 Z 1: 0 I Cnx1Z1: C + i C e x 1 X 1f C 1 e x 1 Y 1

(2.2)

For the minimum of six control points, twelve equations are produced for each view. As there are more equationsthan unknowns, the DLT parameters camera

26

PAYTON CARLJ.

areobtainedbysolvingtheequationsusingaleast-Squarestechnique(Miller, the same M.Lu,rghii.,, 1980).'!fith. the.DLT parameters obtained, *;;;;"J of any co-ordinates three-dimensional to ott"i.t the .q-"u,io", can then U. "r.a provided the two-dimensional co-ordinates of the marker in the obiec, ,f".., marker are known from at least two of the cameras' analysisthe biomechwhen setting up the equipment for three-dimensional below' anist should follow ih. tt.pt in the sub-sections Mount the comeros on stoble tripods ond avoid ponning is tor the cameras to The standard approach in a three-dimensionalanalysis remainStationaryaStheperformermovesthroughtheirfieldofview.ThreecamerasJe.g'.Yu et al', 1993; dimensional filming t..h.riq,r., involving panning have punni,tg and tilting cameras (e'g' Yeadon' L989) Yanai et at., 1996) ".tJ methods these As path' long a occurs over been used when the p.,fot"tu"tt to the global co-ordinate system, a number relative ,iorring cameras the involve each video image to correct for of fixed referencemarkers f,uu. ,o be digitised in method of establishing the changing orientation of the camerui. Att alternative developed by Peak was and tilting video cameras the orientation of p;; tripod instrumented of use performance 1...fr.ologi.s" Inc. This ]tuolu., the positions angular the sense to h;;J, each equipp.d #th two oprical encoders, of the cameras. l1nlmolks Position the cameros for optimum viewing of body . landmarks of interest (e'g' body Grear care should be taken to ensufe that the two cameras for the duration of segment endpoints) ,..n"1., in view of at least can result in the analyst having the activity. Inappropriately positioned cameras which will inevitably movement' stagesof the io g,r.r, li-b poritio* "itt*"1" motion analysis video Many .oipro.rrlr. the accuracy of the co-Jrdinate data' co-ordinates of the predict can offer an interpolation function that p.og'ru--., only be used should option This i^Ulay f""amark that has'becomeobscured. four or five than more no for in situations where the landmark is concealed during that minimum) or (maximum images and is not reaching a turning point period. oll comeros Ensure control points ore visible to ond recorded by . must be clearly visible parameters ,rr.J,o compute the DLT The control points .srhen in using a calibration frame, such as the one shown to .".h ."-",". frame the of rear the. at poles the Fig";. 2.5, care should b. tluk"tt to avoid (or by the tripod). A good contrast L.i"g-"ur."red by those in the foreground "backgroo.d it also essential' It is advisable the between the .o.rtrol point, "nd the start and end of the data collection session' to record the control'points at of the cameras is accidentally This will allow the urrulyrt to recalibrate if one moved slightly during the session' co-ordinate system Align the performonce with the axes of the globol (ISB) recommends that, where The International iociety of Biomechan-ics in. gait, the X-axis examp-le for there is an obvious direction of progression, oftheglobalco.ordinate,y,,..benominallyalignedwiththisdirection.

They propose the being directed vert

Moke provision for Ideally, the two or are synchronised. the camera shutte obtained from ea section in this cha During filmir or strobe light th to confirm that tl corresponds tempc Failure to fulfil d co-ordinates.

digitisin Video

The process of obt on the performe or manually. Mc SIMI'Motion) no markers affixed t< option for the use the performer, e.g automatic trackin in environments v filming outdoors c Manual digit identify and mark will inevitably intr data (seeChapter be kept to an acce manually digitisin r t

The same c consistency( Only ever ul anatomical I musculo-ske Great care s points. Any the co-ordin On comple errors fall w the volume r image and tl

USING VIDEO 27 ANALYSIS MOTION They propose the use of a right-handed co-ordinate system, with the Y-axis being directed vertically and the Z-axis laterally. Moke provision for shufter synchronisation ond event synchronisotion Ideally, the two or more cameras should be gen-locked to ensure their shutters are synchronised. Where this is not possible, the time lag between each of the camera shutters must be determined so the two-dimensional co-ordinates obtained from each camera view can be synchronised (see Video Cameras section in this chapter). During filming, it is useful to activate an event marker e.g. an LED of strobe light that is visible to all cameras. Such a device can be used to confirm that the first video image digitised from a given camera view correspondstemporally to the first image digitised from all other camera views. Failuri to fulfil this requirement will result in erroneous three-dimensional co-ordinates.

ne ny he ch-

to ee-

)3;
lve

ils ber for ing :ak nd 0ns

digitising Video
The process of obtaining two-dimensional co-ordinates of specified landmarks on th. performer, from a video record, may be achieved automatically or manually. Most video motion analysis systems (e.g. APAS, Qualisys, SIMI'Motion) now include software that can automatically track passive markers affixed to the performer. While this facility is clearly an attractive option for the user, it is not always possible or practical to place markers on the performef, e.g. during a sports competition. Even where this is possible, automatic tracking of passive markers can still be problematic, particularly in environments where the contrast level of the marker is variable, e.g. when filming outdoors or underwater. Manual digitising of a video record requires the biomechanist to visually identify and mark the anatomical sites of interest, ftame-by-ftame. This process will inevitably introduce some systematic and random errors to the co-ordinate data (seeChapter 7 for more detail). Sfith attention to detail, these errors can be kept to an acceptablelevel. The following points should be considered when manually digitising a video sequence: o . The same operator should digitise all trials in the study to ensure consistency (reliability) between trials. Only ever use skin-mounted markers as a guide. Consider carefully the anatomical landmark being sought. A sound knowledge of the underlying musculo-skeletal system is essentialhere. Great care should be taken when digitising the scaling object or control points. Any measurement error here will introduce a systematic error in the co-ordinate data. and in all variables derived from these. On completion of a 3D calibration, check that the 3D reconstruction errors fall within acceptable limits. These errors will depend mainly on the volume of the object-space being calibrated, the quality of the video image and the resolution of the digitiser. As a guide, Sanderset al',2006,

e.g. rof ring rbly ysis sof rsed five that

;ible nin
ame lrast iable ;ion. tally

,here -axis tion.

28

CARLJ. PAYTON reported mean RMS reconstructionerrors of 3.9 mm, 3.8 mm and 4.8 mm for the x, y and z co-ordinates, respectively, of 30 points distributed throughout the large calibration frame shown in Figure 2.6. I representative sequenceshould be digitised several times by the investigator to establish the intra-operator reliability. Inter-operator reliability (objectivity) should also be determined by having one or more other experienced individuals digitise the same sequence (see Chapter 7 and measureAtkinson and Nevill. L998. for more information on assessing ment reliabilitv).

ANDPRESENTING ANALYSING PROCESSING, DATA VIDEO-DERIVED


The video digitising process creates two-dimensional image co-ordinates that are contaminated with high frequency errors (noise). Essentially, what is required next is to: 1) smooth and transform the co-ordinates, so that they are in a form suitable for computing kinematic variables, 2) calculate and display the kinematic variables in a format that allows the user to extract the information required to complete the analysis.

co-ordinotes ond tronsforming Smoothing


There are various smoothing methods that can be used to remove the high frequency noise introduced by the digitising process;these fall into three general categories: digital filters, spline fitting and fourier series truncation (Bartlett, 1997a). Failure to smooth co-ordinates sufficiently will lead to high levels of noise in any derived kinematic variables, particularly acceleration. Oversmoothing of the co-ordinates will result in some of the original signal being lost. Selecting the correct smoothing factor, for a given set of co-ordinates, is therefore critically important. Chapter 7 provides a detailed discussion of smoothing methods and presents some practical guidelines for their use. The transformation of image co-ordinates to real world co-ordinates is necessary before any analysis can be undertaken. Procedures for achieving this were discussedearlier in this chapter.

body segment para from cadavers, ge imaging devices(Bz in Robertson, 200 data that closely n analysed. The linear di dimension (e.g. r co-ordinate of that linear displacemen Pythagoras' theore dimensional co-orc relative (e.g. joint r the angle of a seg simple to interpre system have been t three-dimensionalc most common me in biomechanics ar A detailed discussi Linear and an second time deriva These derivatives c method) or analyti functions). As wid and acceleration found.

Anolysing ond p

voriobles ng kinemotic Colculoti


The sport and exercise biomechanist is often interested both in the movement patterns of individual body segments, for example in throwing and kicking' and in the overall motion of the performer's centre of mass, for example in a sprint start. Computation of the mass centre location requires a linked-segment model to be defined, and the mass, and mass centre locations, of individual body segmenrs to be determined. Three general methods are used to obtain

In any biomechani determined by the variables of intere as this will influer high-speed video). deterministic model movement paramet literature. There are a n video analysis and presentation format of the information common methods o peak joint angles)a the focus of the ana plots and angle-ang in sport and exercis

USINGVIDEO29 MOTION ANALYSIS n d


S-

ty er rd
e-

rat is rin lhe ton

righ eral lett, vels vereing Ites, nof The safy


fi/efe

body segment parameters: regression equations based on measurementstaken from cadavers, geometric modelling of the body segments, and the use of imaging devices (Bartlett, 1'997a). More detail on these methods can be found in Robertson,2004. The biomechanist should seek to use segmental inertia data that closely match the physical characteristics of the participants being analysed. The linear displacement of a body landmark (or mass centre) in one dimension (e.g. r direction) is defined as the change in the relevant scaled co-ordinate of that landmark (Ar) during a specified time period. Resultant linear displacements in two or three dimensions are easily calculated using Pythagoras' theorem. Two-dimensional (planar) angles are obtained from twodimensional co-ordinates using simple trigonometry. These angles may be relative (e.g. joint angles formed by two adjacent segments) or absolute (e.g. the angle of a segment relative to the vertical). Planar angles are relatively simple to interpret, once the angular conventions adopted by the analysis system have been established. The calculation of relative (joint) angles from three-dimensionalco-ordinates is more complex, as is their interpretation. The most common methods used for calculating three-dimensional joint angles in biomechanics are the Euler and Joint Co-ordinate System (JCS) methods. A detaileddiscussionof these methods is provided by Andrews,1995. Linear and angular velocities and accelerationsare defined as the first and secondtime derivatives of the displacement (linear or angular), respectively. These derivatives can be computed either numerically (e.g. finite difference method) or analytically (if the data have been smoothed with mathematical functions). As with displacement, the orthogonal components of velocity and acceleration can analysed separately, or their resultants can be found.

doto video-derived ond presenting Anolysing


In any biomechanical analysis, the selection of dependent variables will be determined by the aim of the study. It is important that the biomechanical variables of interest are identified before undertaking the data collection, as this will influence the methodology used (e.g. 2D vs' 3D; normal vs. '!(hen analysing a sport or exercise activity, the use of high-speedvideo). deterministicmodels (Hay and Reid, 1982) can help to identify the important movementparametefs, as of course can reference to the appropriate research literature. There are a number of ways of presenting the kinematic data from a video analysis and it is for the individual to decide on the most appropriate tion format. This will be dictated mainly by the intended destination the information (e.g. research journal, athlete feedback report). The most methods of presenting kinematic data are as discrete measures (e.g. joint angles) and as time series plots (e.g. hip velocity vs. time). Where focus of the analysis is on movement co-ordination, the use of angle-angle and angle-angular velocity (phase)plots is becoming increasingly popular soort and exercise biomechanics.

ment king, :ina iment 'idual ,btain

30

CARL J. PAYTON

REPORTING A VIDEO MOTION ANALYSIS STUDY


The biomechanist should consider including some or all of the following information when reporting a video-based study.

ACKNOWL

I would like to t photographs in thir Mark Johnsonfor

Porticiponts
. o o ' Participant details (age, height, body mass, trained status etc.); Method of obtaining informed consent (verbal or written); Nature of the warm-up and familiarisation; Type of clothing worn, type and position of skin/other markers and the method of locating body landmarks.

REFEREN

Video recording
o . o o r o camera and lens type (manufacturer and model) and the recording medium, format and resolution (e.g. HD 7Z0i on to miniDV tape); Camera settings (frame rate, shutter speed, iris (f-stop) setting); Position of camera(s) relative to the movement being recorded and the field width obtained from each camera (a diagram is useful here); Method used to synchronise the cameras with each other (and with other data acquisition systems if used); Details of lighting (e.g. position of floodlights); Dimensions of 2D scaling object(s) or 3D performance volume (including number and location of control points).

Videodigitising
' . . . Digitising hardware and software (manufacturer and model/version): Resolution of the digitising system; Digitising rate (this may be less than the camera's frame rate): Model used (e.g. 15 point segmental).

Processing, onolysing ond reporting


a O a a a

a a

Algorithm used to obtain the 3D co-ordinates; Method used to smooth/filter the coordinates; Level of smoothing; Method used to obtain the derivative data (e.g. numerical, analytical); Source of segment inertia data used to calculate e.g. the whole body mass centre or moment of inertia: Definitions of the dependent variables being quantified, including their SI units; Estimation of the measurement error in the calculated parameters; Level of inter- and intra-observer reliability of the calculated parameters.

Abdel-Aziz, Y.I. and tor co-ordinates in American Societyc FallsChurch,VA: Andrews, J.G.(1995 three-dimensiona I Three-dimensiona Atkinson, G. and Ne error (reliability)ir 2r7-238. Bartlett, R.M. (199 E & FN Spon. Bartlett, R.M. (ed.) (1 Leeds: BritishAsso Brewin, M.A. and Ke techniques for plan Challis, J., Bartlett,F R.M. Bartlett(ed.) I British Association , Chow, I,U/. (1993)' characteristics of th 9:149-159. Gervais, P., Bedingfie and Kuiper, D. (1 CanadianJournal o Greaves, J.O.B.(199 P. Allard, LA.F. Sto Mouement,Champ Hay, J.G. and Reid,. Motion, Englewoo 'l Kerwin,D.G. (1995) (ed.) Proceedings ol Sport and Exercise S Martin, T.P. and Pon perspective error',R Miller, D.I. and Nels Philadelphia, PA: Le Miller, N.R., Shapiro, I tial kinematic param data', Journal of Bio

USINGVIDEO3I MOTIONANALYSIS

ACKNOWLEDGEMENT
ng I would like to thank Ed Parker for his help in preparing some of the advice.I would also like to thank photographs in this chapterand his technical footagefor Figure2.4. video providing high speed Mark Johnsonfor

REFERENCES
the from comparaY.I. and Karara,H.M. (L971'l'Directlinear transformation Abdel-Aziz, in in closerange photogrammetry', into object spaceco-ordinates tor co-ordinates American Societyof PhotogrammetrySymposiumon Close RangePbotogrammetry, Falls Church, VA: American Societyof Photogrammetry. 'Euler'sand Lagrange's for linked rigid-bodymodelsof equations Andrews, J.G. (1995) human motion', in P. Allard, I.A.F' Stokesand J-P. Blanchi (eds) three-dimensional analysisof buman mouement,Champaign,IL: Human Kinetics. Tbree-dimensional 'Statistical measurement methodsfor assessing Atkinson,G. and Nevill, A.M. (1998) error (reliability) in variablesrelevantto sports medicine', SportsMedicine,26(4); 21.7-238. Bartlett, R.M. (t997al Introduction to Sports Biomechanics,1st edn, London: E & FN Spon. Analysisof Mouementin Sportand Exercise, R.M. (ed.)(1997b)Biomechanical Bartlett, Sciences. of Sport and Exercise British Association Leeds: 'Accuracyof scalingand DLT reconstruction Brewin,M.A. and Kerwin, D.G. (2003) L9:79-88. for planar motion analyses', techniques Journal of Applied Biomechanics, 'Image-based motion analysis',in Challis,J., Bartlett, R.M. and Yeadon, M. (t997) in Sportand Exercise,Leeds: Analysis of Mouement R.M. Bartlett(ed.)Biomecbanical Sciences. of Sport and Exercise BritishAssociation kinematic techniqueto obtain selected Chow, J.W. (1,993)'A panning videographic of the stridesin sprint hurdling', Journal of Applied Biomechanics, characteristics 9:1.49-1.59. 8.V.,'Vfronko, C., Kollias,I., Marchiori, G., Kuntz,J.,'Way,N. P., Bedingfield, Gervais, from panned cinematography', and Kuiper, D. (1989) 'Kinematic measurement 107-1"1'1. Sciences,14: Canadian Journal of Sports 'Instrumentation in systems', three-dimensional in video-based Greaves, J.O.B. (1995) Analysisof Human and J-P. Blanchi(edslThree-dimensional P. Allard, I.A.F. Stokes IL: Human Kinetics. Champaign, Mouement, Hay, J.G. and Reid, J.G. (1982) Tbe Anatomical and MecbanicalBasesof Human Motion, Englewood Cliffs, NJ: PrenticeHall. in J. \Tatkins videodigitisingsystem' Kerwin,D.G. (1995)'Apex/Targethigh resolution (ed.l Proceedingsof the SportsBiomechanicsSection of the British Association of Sciences. of Sportand Exercise BritishAssociation Leeds: Sciences, Sportand Exercise M.B. (1974\ 'Mathematicalcorrectionfor photographic Martin, T.P. and Pongrantz, perspective ch Quarterly, 4 5':378-323. error', Resed.r Approach. of Sport: a Research Miller, D.I. and Nelson, R.C. (1973) Biomechanics PA: Lea & Febiger. Philadelphia, 'A for obtainingspaR. and Mclaughlin, T.M. (1980) technique Miller,N.R., Shapiro, from cinematographic systems of biomechanical of segments tial kinematicparameters chanics,'J-3 t 535- 547. data', Journal of Biome

ling

the ther

ding

'l);
InASS

eir SI

.eters.

32

J. PAYTON CARL

'Body segment parameters', in D.G'E. Robertson, Robertson, D.G.E. (2004) Methods G. Kamen and S.N. Whittlesey(eds)Research G.E. Caldwell, J. Hamill, IL: Human Kinetics. Champaign, in Biomechanlcs, S., McCabe' C.' Naemi' R., ConnaboY'C., Li, S., Scott, G. R., Psycharakis, Sanders, of the art: Applications State A. (2006)'Analysisof swimmingtechnique: and Spence, 2:20-24' supl. Sport Sciences,6, and implications',Portuguese Journal of 'Photographicanalysisof movement',in D.!7. Grieve,D.I. Miller, (1975) G. Smith, D. Mitchelson,J.P.Paul and A.J. Smith (eds)Techniques for the Analysisof Human Mouement,London: LepusBooks. 'Some considerations when applying the Williams, A.M. and Ericsson,K.A. (2005) 24: 283-307. Science, Mouement Human in sport', performance approach expert 'Winter, and Motor Control of HurnanMouement,2ndedn, D.A. (1990)Biomechanics New York: Wiley. 'Three-dimensional videography of Yanai, T., Hay, J.G. and Gerot, J.T. (t9961 673-678. Biomechanics,29: of periscopes', panning with swimming Journal data on ski jumping Yeadon.M.R. (1989) 'A method for obtainingthree-dimensional \ 238-247. Biomechanics, as' usingpan and tilt camer , International Journalof Sports video data', digitised (t999)'A for synchronising method Yeadon,M.R. and King, M.A. 983-986. 32: Biomechanics, of Journal yu, B., Koh, T.J. and Hay, J.G. (1993)'A panningDLT procedure for three-dimensional 26: 741'-751. videography', Journal of Biomechanics,

l,' ,iiiE :1:

CHAPTER 3

, ; :r*riE

MOTTO ON-LIN

ClareE.Miln

INTRODUC

Biomechanicsis a of three-dimensio motion analysisis cise. As with all to opefator to get thr not only the tech quality data, but i our knowledge of biomechanics, res of furthering our prevention of inju by technology, co between the manl rationale. In such technology at the On-line mot to research quest of these studies i of injury or iden sustaining an inju that remaining in and being able t< running and runn popular recreatio sport. Running is its repetitiveness are subtle and c( biomechanical chr