Real handball goalkeeper vs. virtual handball thrower

Benoit BideauR ichard KulpaSt ephane MenardaisdaggerLaet it ia FradetFranck Mult on daggerPaul DelamarcheBruno Arnald i daggerBenoitBideauuhbfrLaboratoire de Physiologie et de

Biomecanique de lrsquoExerciceMusculaire UFR-APSUniversite de Rennes 2

Av Charles Tillon35044 RENNES FRANCE

daggerSIAMES project IRISA

Campus de Beaulieu35042 RENNES FRANCE

Presence Vol 12 No 4 August 2003 411ndash421

copy 2003 by the Massachusetts Institute of Technology

Real Handball Goalk eeper v sVirtual Handball Throw er

A bstract

Virtual reality offers new tools for human motion understanding Several applications

have been widely used in teleoperation military training driving and ying simula-tors and so forth We propose to test if virtual reality is a valid training tool for thegame of handball We focused on the duel between a handball goalkeeper and a

thrower To this end we dened a pilot experiment divided into two steps an ex-periment with real subjects and another one with virtual throwers The throwersrsquomotions were captured in order to animate their avatar in a reality center In this

paper we focused on the evaluation of presence when a goalkeeper is confrontingthese avatars To this end we compared the goalkeeperrsquos gestures in the real andin the virtual experiment to determine if virtual reality engendered the same move-

ments for the same throw Our results show that gestures did not differ betweenthe real and virtual environment As a consequence we can say that the virtual en-vironment offered enough realism to initiate natural gestures Moreover as in real

games we observed the goalkeeperrsquos anticipation to allow us to use virtual realityin future work as a way to understand the goalkeeper and thrower interactionsThe main originality of this work was to measure presence in a sporting application

with new evaluation methods based on motion capture

1 Introd uct ion

Studying human motor control still remains a challenging task Indeednatural motion control is the consequence of many parameters (biomechanicalphysiological psychological sociological et cetera) Among all these parame-ters only a few of them can be captured Hence experiments that consist ofmeasuring only a subset of these parameters may ignore crucial ones The mainproblem is to ensure that the conditions are exactly the same between severalseparated experiments Without this constraint no generality about motioncontrol can be deduced and the large standard deviations are considered asnoise whereas it may be due to different experimental conditions

Let us consider now a handball goalkeeper that tries to stop balls One ofour aims was to identify the elements used by the handball goalkeeper to selecta strategy when he is confronting a thrower Real experiments raised the ques-tion of reproducibility Indeed nothing ensures that two throws are exactlyidentical even those performed by a unique thrower who does his or her bestto throw exactly the same way Several external parameters may interfere withthe experiments such as a small change in the motion coordination of thethrower the behavior of the crowd and the background noise In conclusionwe cannot link the handball goalkeeperrsquos strategy to a unique situation if we

Bideau et al 411

cannot eliminate such random phenomena Virtual real-ity is a promising tool to overcome these limitationsbecause it makes it possible to ensure reproducibility

Nevertheless using virtual reality instead of conduct-ing real experiments raises new questions First we haveto take presence into account Presence denotes thesubjective feeling of ldquobeing thererdquo and was mainly stud-ied through behavioral analyses (Schuemie Van DerStaaten Krijn Van Der Mast CAPG 2001) Hence inour case did the subject really react as in the real worldTo this end the virtual environment needs to be asclose to a real scene as possible Synthetic buildingsareas characters and motions have to be consideredThe role of the quality of the geometric model has beenpointed out by Hodgins OrsquoBrien and Tumblin (1998)Second in addition to graphical realism the syntheticcharacters that inhabit the virtual environment may actas real actors do Thalmann (1996) categorized fourkinds of virtual characters

c Avatars act exactly as the user doesc Guided actors are driven by users via the concept of

metaphorsc Autonomous agents have their own behavior They

perceive their environment and can interact with itc Perceptive and interactive agents are autonomous

agents who are aware of other actors and can com-municate with them

Moreover we also have to consider the interactionsthat are necessary to make the real subject interact withthe virtual world Several kinds of interactions may oc-cur between these agents Noser Pandzic Capin Mag-nenat-Thalmann and Thalmann (1996) experimentedwith an interaction between a real tennis player (repre-sented by his avatar) and a virtual one The virtualplayer was a perceptive and interactive agent driven by abehavioral model However the avatar was as depictedonly a part of a human body composed of an arm and aracket The captured motion of the real tennis player isreplayed on this virtual arm without taking any motiondetails into account No attention was paid to the ef-fects given to the ball and to the complex player dis-placements all over the court Molet et al (1999) made

two real players interact with each other in a shared en-vironment via the VLNET network (Capin PandzicNoser Magnenat-Thalmann amp Thalmann 1997) Thetwo players saw their own avatar (only the arm and theracket) playing with a synthetic human-like gure As inthe previous work the user was not able to play tennisas in the real world This kind of virtual game cannot beapplied directly to study real sporting motions per-formed by high-level athletes Other applications werededicated to evaluating and training sportsmen Forexample virtual reality was used to test a collective sportstrategy by immersing a coach in a simulated game (Me-toyer amp Hodgins 2000) In this game a coach gaveorders to synthetic players while a behavioral modeldrove the opponents Nevertheless the simulated be-havior was again not compared to real situations and thecoach did not act as in real games but used metaphorsto drive his team On the contrary bobsleigh driverswere asked to train on a simulator for the winter Olym-pic games (Huffman amp Hubbard 1996) In this studythe simulator was designed to make the drivers behaveas in the real world However no analysis was per-formed to verify if the driver really acts as in the realworld

To sum up previous work on virtual reality was gen-erally based on metaphors to animate avatars and todrive the virtual environment We propose a set of ex-periments to verify if the motions performed by a sub-ject immersed in a virtual environment are similar to theones performed in the real world Task performancerelated to the level of presence was previously studied(Slater Linakis Usoh amp Kooper 1996) In this earlierwork behavioral analyses were carried out without ac-counting for the subjectrsquos gestures In sport gesturesare directly linked to performance so evaluating thegesture is necessary We performed a two-step processFirst we captured the motion of a handball thrower andgoalkeeper during a real duel Then we replayed thethrowerrsquos motion in a virtual environment and we veri-ed if the goalkeeper reacted as in the real world by cap-turing his motion In the second step we modeled asynthetic thrower that had to replay the captured mo-tions of real handball throwers In that sense it could

412 PRESENCE VOLUME 12 NUMBER 4

be viewed as an avatar that acted as a real subject butnot in real time The synthetic actor was designed to beable to perform other motions (such as those computedby a model) but only the restitution of captured mo-tions was used in this work The quality of the avatarmotion was essential to ensure presence

To this end we chose to replay captured motionswhile accounting for constraints such as ensuring footcontact without sliding adapting the motion to the syn-thetic skeleton and so on We also wished to modifyslightly the original motion to investigate the conse-quences of each change on the goalkeeperrsquos behaviorAs a consequence we chose to use motion warping andblending techniques Motion warping was experimentedwith by modeling the trajectories in the temporal (Wit-kin amp Popovic 1995) or frequency domain (UnumaAnjyo amp Takeuchi 1995) The main drawback of thefrequency domain parameterization was the lack of theresulting motion controllability Indeed changing theweight of a harmonic is not intuitive and drives us to atrial-and-error repetitive process Changing controlpoints or adding space-time constraints (Witkin amp Kass1988) is more intuitive Captured motions have to becorrected to be used with such a method For examplenoise has to be ltered out and anatomical correctionsperformed (Molet Boulic amp Thalmann 1996 Boden-heimer Rose Rosenthal amp Pella 1997) For our appli-cation procedural animation (Zeltzer 1982 BoulicMagnenat-Thalmann amp Thalmann 1990 Bruderlin ampCalvert 1996) and dynamic simulation (Arnaldi Du-mont Hegron Magnenat-Thalmann Thalmann 1989Hodgins Wooten Brogan amp OrsquoBrien 1995 MultonNougaret Hegron Millet amp Arnaldi 1999) were notappropriate Indeed even for accurate models the syn-thetic motions could not be reliably compared to theones measured in real games

In Section 2 we describe the methods chosen to ani-mate the virtual thrower Section 3 gives informationabout the real and virtual experiments carried out withone handball goalkeeper Finally we conclude by askingif such a method may be useful to better understand theduel between the handball goalkeeper and the throwerand give some perspectives

2 From Mocap to Virtual R ea lity

The rst part of this study is a preliminary experi-ment involving a real handball goalkeeper and throwersHandball is a game confronting two teams of sevenplayers against each other on a 40 3 20 m eld Theaim is to make goals by throwing a ball with the arm ata minimum of 6 m away from a goalkeeper Hence aduel between the thrower and the keeper occurs andraises some strategic questions What kind of informa-tion is used by the goalkeeper to intercept the ball

The aim of this study was to capture the motion of areal handball thrower together with the correspondinggoalkeeperrsquos motion As a consequence we were able toanalyze the goalkeeperrsquos reaction in front of severalhandball throws we identied space-time constraintsthat link the thrower (for each throw) and the goalkeep-errsquos behavior To this end we used an optoelectronicmotion capture system Vicon370 composed of seveninfrared cameras set up at a frequency of 60 Hz (SeeFigure 1)

The cameras were placed to cover a 12 3 6 m eld ofmeasurement To cover this space and to better capturethe action of the thrower and the goalkeeper we placedthe cameras along a circle all around the playing areaThe subjects were tted with 26 circular reective mark-ers to precisely reconstruct the 3D orientation of eachsegment (See Figure 2)

To animate a synthetic skeleton with the resulting

Figure 1 Handball thrower and goalkeeper

Bideau et al 413

trajectories we performed a set of post-processing Forour study we used a 29-DOF (including three rotationsaround the shoulder one around the elbow nonearound the wrist) geometric model Among otherthings optical motion capture systems imply dealingwith occlusions (missing points) Hidden markers arenot reconstructed and the resulting 3D trajectories ex-hibit holes Of course before animating a human-likegure we had to retrieve these missing points We useda global framework to animate virtual handball playersthanks to motion capture (Menardais Multon amp Ar-naldi 2002) This framework generated trajectorieswithout occlusions that could be played in real timeMoreover the captured trajectories were adapted to thevirtual playerrsquos morphology Indeed the real subjectand the virtual one generally had different sizes andmorphologies The global framework also enabled us toadapt the original motion to the virtual actor while en-suring foot contact with the ground

This framework was organized as follows First it en-abled us to load a motion capture le with its corre-sponding original skeleton (linked to the real subjectwho performed the motion) Second we ran the recon-

struction process that recovered the missing pointswhile accounting for anatomical constraints These con-straints were mainly distance constraints between twopoints belonging to the same body segment Third weloaded the skeleton of the virtual player that is then ani-mated The reconstructed trajectories were applied tothis last skeleton by computing the required quaternionsfor each articulation At this step we activated con-straints such as foot contact with the ground and conti-nuity with other compatible motions For example ifthe studied motion m1 had to be sequenced with a runm2 a left foot strike at the end of m1 should corre-spond to a left foot strike at the beginning of m2 Theresulting motion was ltered with a second-order But-terworth lter with a 10 Hz cutoff frequency At theend of this process the motion was stored automaticallyas a set of keyframes that were directly used for com-puter animation

Several captured motions were necessary to animate avirtual handball player running walking and varioushandball throws These throws corresponded to theones performed by the handball player during the pre-liminary experiment As the capture area did not coverall the motion from the start to the ball release includ-ing the running phase we captured some of these mo-tions separately from the throw itself Consequently themotions were sequenced in real time by the animationengine In conclusion we paid attention to the transi-tion between these elementary motions To this endthe animation engine was able to synchronize severalmovements This engine was embedded in a virtual real-ity platform GASP (general and simulation platform)(Donikian Chauffaut Duval amp Kulpa 1998) GASP isa software framework that enables communication andinteraction between entities belonging to a virtual envi-ronment This communication can be achieved in twoways

c dataow Each entity owns its own inputs outputsand control parameters The inputs of one modulecan be dynamically plugged to the outputs of an-other one during the simulation For examplewhen the ball was attached to the handball playerrsquoshand the output coordinates of his hand were

Figure 2 Location of the reective markers on the subject


transmitted to the ball whereas this link was brokenafter ball release

c events GASP offers services to manage signalsevents and event listeners Events were used tomodel the ball release and the change of motion(for example the move from running to throwingwhen entering the 9 m area)

Our animation engine used three entities as depictedin Figure 3

c the thrower This module represented the virtualthrower First it dealt with his behavior To thisend it managed the scheduling of his motions Sec-ondly this entity managed the 3D animation of thevirtual thrower It used the captured motion to getthe quaternions for each articulation Moreover ithandled the transition between them to have a real-istic animation

c The throw congurer This entity dealt with the in-teraction between the new motion of the throwerand the trajectory of the ball Each time thethrower played a motion this module accessed thedatabase to get the following parameters right

c active In our application this Boolean speciedwhether the new motion of the thrower had aneffect on the ball trajectory It was the case dur-ing a throw or during a run when the ball is held(The parameter is set to ldquotruerdquo) On the con-trary after a throw the thrower ran while the ballwas following its own trajectory In that case thisparameter was set to ldquofalserdquo and all the otherparameters were ignored

c thrown At the beginning of the motion (if theldquoactiverdquo parameter was set) the ball was in thehand of the thrower This parameter was a Bool-ean that specied whether the ball had to bethrown Indeed during the run before thethrow the virtual thrower kept the ball in hishand In this case (parameter set to ldquofalserdquo) thefollowing three parameters were not used

c ball release event During a throwing motion itcorresponded to the time of ball release

c ball speed It was the ball speed during its yingphase after release

c ball destination It was the place where the ball en-tered the goal

c the ball handler After getting all the parametersfrom the previous module this entity managed the3D trajectory of the ball During a motion in whichthe ball was in the hand of the thrower this moduleobtains the wrist position of the thrower The posi-tion of the ball corresponded to the wrist one

3 R eal Handball Goalk eeper v s VirtualHandball Play er

In the second part of the experiment we placedthe real handball goalkeeper in a virtual stadium (seeFigure 4) to play against a virtual thrower (Figure 5)We used a reality center comprising a SGI Onyx2 In-niteReality with three pipes and three Barco 1208Svideoprojectors used on a cylindrical screen (with a ra-dius of 380 m a height of 238 m and a 135deg eld ofvision) To obtain a real goalkeeperrsquos behavior we re-constructed an environment as real as possible by repro-ducing visual landmarks well known in handball Oneof the most important landmarks was the goal itself thatwas physically placed in the reality center This goal wasplaced around the goalkeeperrsquos position that also corre-sponded to the position of the virtual camera First toset up the position of the virtual camera we placed atemporary landmark into the virtual environment It wasplaced at a distance from the virtual goal correspondingto the one between the real goal and the screen Thenwe moved the camera until the landmark intersected thedisplay The remainder of the virtual stadium was mod-eled to t the dimension of a real stadium

Figure 3 Animation engine

Bideau et al 415

Once the real and the virtual world were calibratedwe focused on the movement of the virtual ball whichwas another important parameter that engendered goal-keeper reactions This ball had to be textured as a realball and its trajectory after release had to satisfy the me-chanical laws To this end three parameters were neces-sary the location of the ball release in space the ballspeed at release and the location of the intersectionpoint with the goal To best t the original throw wemeasured the actual ball speed for each studied throwwith a radar gun (from Stalker Radar) In the real exper-iment the ball release was determined with the motioncapture system that provided us with the location of theball at release for all the studied throws Finally we sub-divided the goal into several areas including the upper-left corner and the upper-right corner In the realworld the players shot only on three of these areas Theother areas were used to create new situations by chang-ing articially the ball destination for a given studiedthrow The application of this method was to determineif the goalkeeper reacted differently and consequentlyif he took information from only the ballrsquos trajectory orfrom other parameters

To study the different throws and the resulting goal-keeper actions we again used the Vicon370 motioncapture system The reality center and the motion cap-

ture system were not physically synchronized and nostart signal was given to the goalkeeper who naturallyreacted to the virtual thrower actions

For the virtual experiment one goalkeeper partici-pated in this pilot experiment and was equipped withthe markers The subject took his place in the realitycenter inside a goal that was physically placed in theroom We familiarized the goalkeeper with the virtualenvironment and all the equipment used for the experi-mentation (3D glasses markers and the virtual environ-ment) As in the real experiment no particular instruc-tion was given to the subject He only had to react as ina real game without restriction Then we asked the goal-keeper to stop 24 throws that were randomly chosenamong all the available captured throws These capturedthrows were divided into three main categories

c 6 m throw without jumping using four capturedthrows and another faked one (the ball destinationwas articially changed)

c 6 m throw while jumping using four capturedthrows

c 9 m throw without jumping using three capturedthrows

All these throws were played two times randomly toprevent the goalkeeper from recognizing the throw

Figure 4 Virtual stadium Figure 5 Virtual thrower vs real goalkeeper


used Between each throw the subject returned to hisnatural position and waited for the next trial withoutany information about the following trial

4 R esu lt s

The main goal of these experiments was to verify ifthe movements performed by the goalkeeper in the realand virtual environments were similar and thus if vir-tual reality could be used in sport as a training and re-search tool In our study we focused on biomechanicalanalysis To this end we chose to compare the arm andleg displacements in the real and the virtual environ-ment More specically we studied the arm and the legcenter of mass (denoted COM in the remainder of thepaper) displacement in the total body COM referenceframe We selected the arm and the leg COM becausenot only the hands or the feet were used to catch theball but the whole limbs The COM was computed withanthropometrical tables (Winter 1979) To comparetrajectories obtained in the real and virtual environmentwe had to ensure that these data were compatible Tocompare two trajectories we determined an event thatenabled us to synchronize the two motions In a hand-ball throw the ball release could be such an eventHowever as there was no physical synchronization be-

tween the motion capture system and the reality centerthe ball release in the virtual world could not be viewedwith the motion capture system Consequently wechose to synchronize the trajectories according to thebeginning of the goalkeeperrsquos reaction To this end wecomputed the acceleration of the goalkeeperrsquos arm in allthe studied trajectories A peak of acceleration was iden-tied in all the resulting accelerations and represents thebeginning of the goalkeeperrsquos action (See Figure 6)Once this event was detected for all the studied trajecto-ries we selected a time window ranging from 203s to103s around it

We studied 24 throws divided into three main cate-gories from 6 m with and without jumping and from9 m without jumping For each category four ball des-tinations were considered including all the goalrsquos cor-ners As a consequence the goalkeeper was confrontedrandomly several times with the same situation For oneof all these possible situations Figure 7 presents thedisplacement of the arm COM along the z axis (verticalaxis) knowing that this arm was going to intercept theball For each studied situation all the considered COMtrajectories had the same shape playing in a real or avirtual handball environment Hence there was no sig-nicant difference between the motions for the real andthe virtual situations

Table 1 describes the results obtained by considering

Figure 6 Acceleration of arm COM for a motion Figure 7 Center of mass displacement of the right arm along the z

axis

Bideau et al 417

the armrsquos movement In this table the three situationsappear in bold font For each situation we give all thethrows in the real experiment together with their corre-sponding throws in the virtual environment Each throwin the virtual experiment was compared to the real onethrough the initial and nal arm position in the globalCOM reference frame the armrsquos displacement inmeters the difference between the armrsquos displacementin the real and virtual throw in percentage and the cor-relation between the shape of the movements in the realand virtual experiments

Table 2 contains the same kind of information but forthe leg

The worst difference was about 94 6 25 (that is tosay 119 at worst) along the z axis for the armrsquos COMmotion

In addition for some trials we articially modiedthe ballrsquos trajectory to make it reach the opposite cornerwithout changing the handball thrower motion Eachtime the goalkeeper tried to stop the ball on the origi-nal side His repeated mistakes were really interestingbecause they showed that the goalkeeper did not take

Table 1 Kinematic Variations of the Armrsquos Center of Mass Along the Vertical Axis

MotionArm initialposition (m)

Arm nalposition (m)

Displacement(m)

Differencefrom the realaction ()

Correlationcoefcient(R2)

6 mReal 1 0546 0706 016 Virtual 1 0522 6 003 0667 6 000 0145 6 004 94 6 25 098 6 001Real 2 0542 0400 0142 Virtual 2 0528 6 004 039 6 003 0136 6 002 4 6 12 099 6 000Real 3 0536 0661 0125 Virtual 3 0520 6 005 0632 6 004 0112 6 001 104 6 08 097 6 001Real 4 0539 0382 0157 Virtual 4 0525 6 001 0355 6 003 017 6 002 82 6 12 098 6 001

6 m 1 jumpReal 1 0535 0336 0199 Virtual 1 0516 6 001 0330 6 002 0186 6 001 65 6 05 098 6 000Real 2 0543 0722 0179 Virtual 2 0533 6 002 0702 6 005 0169 6 003 56 6 16 099 6 000Real 3 0551 0350 0201 Virtual 3 0525 6 003 0340 6 002 0183 6 000 87 6 02 097 6 001Real 4 0538 0703 0165 Virtual 4 0520 6 005 0672 6 001 0152 6 004 78 6 25 098 6 001

9 mReal 1 0545 0695 015 Virtual 1 0526 6 004 0662 6 003 0136 6 001 93 6 07 097 6 001Real 2 0537 0712 0175 Virtual 2 0517 6 001 0685 6 000 0168 6 001 45 6 1 098 6 001Real 3 0549 0704 0155 Virtual 3 0528 6 003 0670 6 002 0142 6 001 84 6 06 097 6 000


information from the ball trajectory but from thethrowerrsquos motion

5 Discussion

In this paper we presented a virtual reality experi-ment that involved a real goalkeeper and several virtualthrows The goal of this paper was to verify if the goal-keeperrsquos reactions in the virtual environment were simi-lar to those captured in a preliminary real experiment

Usually presence in a virtual world has been quantiedusing questionnaires and statistical analysis (Slater1999) Usoh Catena Arman and Slater (2000) dem-onstrated that such studies are limited for a comparisonbetween real and virtual environments In our study wewere particularly interested in the movements So weused biomechanical analysis Our last goal was to pro-duce a presence index The motions of the virtualthrower were captured during the preliminary experi-ment and adapted to t the virtual throwerrsquos skeletonThe ball was driven by a mechanical model whose in-

Table 2 Kinematic Variations of the Legrsquos Center of Mass Along the Lateral Axis

MotionLeg initialposition (m)

Leg nalposition (m) Displacement (m)


Correlationcoefcient (R2)




Bideau et al 419

puts were the initial position at release the velocity vec-tor norm and the destination in the goal Our resultsshowed that the goalkeeperrsquos movements in the virtualenvironment were similar to those captured in the realexperiment This result was obtained for all the studiedsituations including three different throws each withtwo different destinations

Of course we tested this experiment with only onesubject who played at a national level but the promisingresults encourage us to repeat this experiment with alarger set of goalkeepers and throwers These kinds ofresults are very important for sport applications Indeedthis tool offers new ways of investigation to understandhow goalkeepers react to throws One of the most im-portant points is to reproduce exactly the same situationand verify if the goalkeeper always reacts in the sameway Another important point is to ensure that the syn-thetic motions are realistic enough to engender realisticreactions For our study we used a 29-DOF modelwhose trunk was modeled as a unique body segmentThis simplied model was enough to make the goal-keeper react as he would against a real thrower Never-theless future work will tend to improve the geometricmodel and the skeleton quality (such as adding ngersexion and other rotational joints to the trunk)

We also experimented with a few fake throws a realmotion for which the ball should go to the left cornerwhereas we made the ball go to the other side Thegoalkeeper each time tried to stop the ball on the origi-nal side and was tricked by our modication This was avery important point suggesting that the goalkeepertook information from the throwerrsquos motion and re-acted before the ball was released Indeed it was knownthat at a high playing level goalkeepers do not takeinformation from the gaze of throwers but perhaps fromthe ball trajectory So this rst result was really interest-ing This experiment could not be conducted with realthrowers because changing the ball direction also in-volved changing the movement Hence it was quiteimpossible to verify this hypothesis with real experi-ments

This system provides trainers with new tools to evalu-ate the goalkeeperrsquos performance and training In addi-tion this tool offers a way to train goalkeepers against

future opponents whose motions are designed accord-ing to old recorded games (for example from video-tapes and with the help of an animator)

An improvement on this technique should add thepossibility for the goalkeeper to interact with thethrower To this end we need a real-time motion cap-ture system that would be able to analyze the goalkeep-errsquos reactions in order to change the throwerrsquos strategyIn some cases the goalkeeper anticipates the throwerrsquosaction by voluntarily blocking a side of the goal Thenthe thrower is encouraged to throw the ball at the op-posite side During our virtual experiments the goal-keeper explained that he was tempted to try this strat-egy Nevertheless in this version there was no changein the virtual throwerrsquos behavior Future work will tendto overcome this limitation by coupling to the realitycenter a real-time motion capture system and a simplebehavioral model Adding such a system will also makeit possible to verify if the goalkeeper intercepts the ballby performing a collision detection between the ballrsquosgeometry and the goalkeeperrsquos upper limb We are cur-rently working on such an experiment

Another possible extension of this tool is to offer thetrainer and the goalkeeper a posteriori view of a cameraplaced anywhere in the 3D world For example thecamera could be placed to see at the same time the mo-tions of the thrower and the goalkeeper to observe theinteraction between them Hence it would be possibleto visualize the movements of the goalkeeper for all thesituations Consequently we can conduct the oppositeexperiment by examining how a thrower should react toa goalkeeperrsquos motions for all the studied situations Wecould also use a distributed environment to make tworeal players interact through two distant reality centersas in Noserrsquos experiment (Noser 1996) Such a tech-nique may allow distant players to train each other whilerecording their motion in 3D for a posteriori analyses

A ck now ledgments

This work has been supported by the French Ministry ofSport Youth and the French Olympic Preparation committeeand the Conseil Regional de Bretagne


R eferences

Arnaldi B Dumont G Hegron G Magnenat-Thalmann

N amp Thalmann D (1989) Animation control with dy-namics Proceedings of Computer Animation 89 113ndash123

Bodenheimer B Rose C Rosenthal S amp Pella J (1997)

The process of motion capture Dealing with the data Pro-ceedings of Eurographics Workshop on Computer Animationand Simulation 3ndash18

Boulic R Magnenat-Thalmann N amp Thalmann D

(1990) A global human walking model with real-time kine-matic personication The Visual Computer 6(6) 344ndash358

Bruderlin A amp Calvert T (1996) Knowledge-driven inter-

active animation of human running Proceedings of GraphicsInterface 96 213ndash221

Capin T Pandzic I Noser H Magnenat-Thalmann N

amp Thalmann D (1997) Virtual Human Representationand Communication in VLNET Networked Virtual Envi-ronments IEEE Computer Graphics and Applications Spe-cial Issue on Multimedia Highways 17(2) 42ndash53

Donikian S Chauffaut A Duval T amp Kulpa R (1998)GASP From Modular Programming to Distributed Execu-tion Computer Animation lsquo98 pp 79 ndash87

Hodgins J OrsquoBrien J amp Tumblin J (1998) Perception ofhuman motion with geometric models IEEE Transactionon Visualization and Computer Graphics 4(4) 307ndash316

Hodgins J Wooten W Brogan D amp OrsquoBrien J (1995)Animating human athletics Proceedings of ACM SIG-GRAPH 71ndash78

Huffman K amp Hubbard M (1996) A motion based virtual

reality training simulator for bobsled drivers The engineer-ing of sport 195ndash203

Menardais S Multon F amp Arnaldi B (2002) A global

framework for motion capture Research report INRIA No4360

Metoyer R amp Hodgins J (2000) Animating athletic mo-

tion planning by example Proceedings of Graphics Interface2000 61ndash68

Molet T Aubel A Capin T Carion S Lee E Mag-nenat-Thalmann N Noser H Pandzic I Sannier G amp

Thalmann D (1999) Anyone for tennis Presence Teleop-erators and Virtual Environments 8(2) 140ndash156

Molet T Boulic R amp Thalmann D (1996) A real-timeanatomical converter for human motion capture Euro-graphics Workshop on Computer Animation and Simulation79ndash94

Multon F Nougaret J L Hegron G Millet L amp Ar-naldi B (1999) A software toolbox to carry out virtualexperiments on human motion Computer Animation 16ndash23

Noser H Pandzic I Capin T Magnemat-Thalmann Namp Thalmann D (1996) Playing games through the virtuallife network ALIFE V 114 ndash121

Schuemie M J Van Der Straaten P Krijn M Van DerMast amp CAPG (April 2001) Research on presence in VRCyberpsychology and Behavior 183ndash202

Slater M Linakis V Usoh M amp Kooper R (1996) Im-mersion presence and performance in virtual environ-ments An experiment in tri-dimensional chess Proceedingsof VRST rsquo96 163ndash172

Slater M (1999) Measuring presence A response to the Wit-mer and Singer questionnaire Presence Teleoperators andVirtual Environments 8(5) 560 ndash566

Thalmann D (1996) A new generation of synthetic actorsThe interactive perceptive actors Proceedings of PacicGraphics lsquo96 200ndash219

Unuma M Anjyo K amp Takeuchi R (1995) Fourier prin-ciples for emotion-based human-gure animation Proceed-ings of ACM SIGGRAPH 91ndash96

Usoh M Catena E Arman S amp Slater M (2000) Pres-ence questionnaires in reality Presence Teleoperators andVirtual Environments 9(5) 497ndash503

Winter D (1979) A new denition of mechanical work donein human movement Journal of Applied Physiology 46(1)78ndash 83

Witkin A amp Kass M (1988) Spacetime constraints Pro-ceedings of ACM SIGGRAPH 159 ndash168

Witkin A amp Popovic Z (1995) Motion warping Proceed-ings of ACM SIGGRAPH 105ndash108

Zeltzer D (1982) Motor control techniques for gure ani-mation IEEE Computer Graphics and Applications 2(9)53ndash59

Bideau et al 421

cannot eliminate such random phenomena Virtual real-ity is a promising tool to overcome these limitationsbecause it makes it possible to ensure reproducibility

Nevertheless using virtual reality instead of conduct-ing real experiments raises new questions First we haveto take presence into account Presence denotes thesubjective feeling of ldquobeing thererdquo and was mainly stud-ied through behavioral analyses (Schuemie Van DerStaaten Krijn Van Der Mast CAPG 2001) Hence inour case did the subject really react as in the real worldTo this end the virtual environment needs to be asclose to a real scene as possible Synthetic buildingsareas characters and motions have to be consideredThe role of the quality of the geometric model has beenpointed out by Hodgins OrsquoBrien and Tumblin (1998)Second in addition to graphical realism the syntheticcharacters that inhabit the virtual environment may actas real actors do Thalmann (1996) categorized fourkinds of virtual characters

c Avatars act exactly as the user doesc Guided actors are driven by users via the concept of

metaphorsc Autonomous agents have their own behavior They

perceive their environment and can interact with itc Perceptive and interactive agents are autonomous

agents who are aware of other actors and can com-municate with them

Moreover we also have to consider the interactionsthat are necessary to make the real subject interact withthe virtual world Several kinds of interactions may oc-cur between these agents Noser Pandzic Capin Mag-nenat-Thalmann and Thalmann (1996) experimentedwith an interaction between a real tennis player (repre-sented by his avatar) and a virtual one The virtualplayer was a perceptive and interactive agent driven by abehavioral model However the avatar was as depictedonly a part of a human body composed of an arm and aracket The captured motion of the real tennis player isreplayed on this virtual arm without taking any motiondetails into account No attention was paid to the ef-fects given to the ball and to the complex player dis-placements all over the court Molet et al (1999) made

two real players interact with each other in a shared en-vironment via the VLNET network (Capin PandzicNoser Magnenat-Thalmann amp Thalmann 1997) Thetwo players saw their own avatar (only the arm and theracket) playing with a synthetic human-like gure As inthe previous work the user was not able to play tennisas in the real world This kind of virtual game cannot beapplied directly to study real sporting motions per-formed by high-level athletes Other applications werededicated to evaluating and training sportsmen Forexample virtual reality was used to test a collective sportstrategy by immersing a coach in a simulated game (Me-toyer amp Hodgins 2000) In this game a coach gaveorders to synthetic players while a behavioral modeldrove the opponents Nevertheless the simulated be-havior was again not compared to real situations and thecoach did not act as in real games but used metaphorsto drive his team On the contrary bobsleigh driverswere asked to train on a simulator for the winter Olym-pic games (Huffman amp Hubbard 1996) In this studythe simulator was designed to make the drivers behaveas in the real world However no analysis was per-formed to verify if the driver really acts as in the realworld

To sum up previous work on virtual reality was gen-erally based on metaphors to animate avatars and todrive the virtual environment We propose a set of ex-periments to verify if the motions performed by a sub-ject immersed in a virtual environment are similar to theones performed in the real world Task performancerelated to the level of presence was previously studied(Slater Linakis Usoh amp Kooper 1996) In this earlierwork behavioral analyses were carried out without ac-counting for the subjectrsquos gestures In sport gesturesare directly linked to performance so evaluating thegesture is necessary We performed a two-step processFirst we captured the motion of a handball thrower andgoalkeeper during a real duel Then we replayed thethrowerrsquos motion in a virtual environment and we veri-ed if the goalkeeper reacted as in the real world by cap-turing his motion In the second step we modeled asynthetic thrower that had to replay the captured mo-tions of real handball throwers In that sense it could











Bideau et al 413






















Bideau et al 415












4 R esu lt s






axis

Bideau et al 417







Arm nalposition (m)

Displacement(m)








5 Discussion











Bideau et al 419








A ck now ledgments



R eferences



































Bideau et al 421










Bideau et al 413






















Bideau et al 415












4 R esu lt s






axis

Bideau et al 417







Arm nalposition (m)

Displacement(m)








5 Discussion











Bideau et al 419








A ck now ledgments



R eferences



































Bideau et al 421






















Bideau et al 415












4 R esu lt s






axis

Bideau et al 417







Arm nalposition (m)

Displacement(m)








5 Discussion











Bideau et al 419








A ck now ledgments



R eferences



































Bideau et al 421












4 R esu lt s






axis

Bideau et al 417







Arm nalposition (m)

Displacement(m)








5 Discussion











Bideau et al 419








A ck now ledgments



R eferences



































Bideau et al 421







Arm nalposition (m)

Displacement(m)








5 Discussion











Bideau et al 419








A ck now ledgments



R eferences



































Bideau et al 421


5 Discussion











Bideau et al 419








A ck now ledgments



R eferences



































Bideau et al 421








A ck now ledgments



R eferences



































Bideau et al 421

R eferences



































Bideau et al 421

Real handball goalkeeper vs. virtual handball thrower

Documents

Transcript of Real handball goalkeeper vs. virtual handball thrower