Football is a challenging research domain. Each match involves 22 players
Download 179.11 Kb. Pdf ko'rish
|
Football
- Bu sahifa navigatsiya:
- Detection of real-time patterns in sports interactions in football Gudberg K. Jonsson, Sigridur H. Bjarkadottir, Baldvin Gislason
- Figures 4a et 4b
- Figure 8
37 Introduction Football is a challenging research domain. Each match involves 22 players d em on s tr a tin g c o lla b o ra ti ve b e h a vi ou r th a t r eq u ir e s s p e c ifi c r o le s in an adversarial, uncertain, and dynamic environment. The behaviour of players and the decision making processes can range from the most simple reactive behaviours, such as running towards the ball, to complex reasoning that take in to a cc o u n t th e b eh a v io u r a n d p e rc e iv e d s tr a te g ie s o f te a m-ma te s a nd opponents (Jonsson, 1998). In the pursuit of generating quantitative information on performance sport researchers have traditionally used frequency of event occurrence as their index of performance e.g. the analyst has recorded how many passes have been made from particular playing zones or how many times possession has been lost (Jonsson et al., 2000). In essence the analyst has been answering the question "how many times did ’x’ occur?". However frequency of event occurrence has been shown to be an inadequate index of performance that ca n n o t d i ff e re n ti at e b et we e n e ffe c t iv e p e r fo rm a nc e s (B o r rie a n d J on e s , 19 9 8). I f on e a cc e pts th e a rg u men t th at s po rt p er fo rman c e co n sis ts of a complex se ries of inte rrelations hips b etwe en a vas t arra y o f pe rformance v ar ia b l e s th e n s imp l e fr e q u en c y d a ta c a n o n ly e v er p r o vi d e a re l a tiv e l y superficial view of performance. If performance analysis is to continue to advance understanding of sports pe rforma nce then it mus t fin d be tter me thod s of co llec ting a nd a naly zing match analysis data. The purpose of this paper is to introduce and explain a new data analysis method that has the potential to make a significant contri- bution to analyses of sports performance. Data from preliminary studies of football performance are also presented to show the potential outcome from the analysis process. T-pattern detection and analysis T h e a n a ly s is a p p ro a c h p re s e n t e d i s b a s e d o n a p ro c e s s k n o wn a s T- pattern detection which allows the detection of the temporal and sequential structure of a data set. The method has been developed, outside of sport, on the assumption that complex streams of human behaviour have a temporal/ sequential structure tha n can not be fully detected through un aid ed ob ser- v a t i o n o r w i t h t h e h e l p o f s t a n d a r d s t a t i s t i c a l a n d b e h a v i o u r a n a l y s i s me th ods. Give n tha t ob servation al reco rds o f hu ma n be havio ur, including sport performance analysis, have both a temporal and sequential structure an analysis tool that can describe this structure will enhance understanding of the behaviour (s) being studied. A generic observational software package called The me has be en spec ifically deve lo ped to o peration alis e T-pa tte rn detection as an analysis process (Magnusson, 1996, 2000).
Gudberg K. Jonsson, Sigridur H. Bjarkadottir, Baldvin Gislason, Andrew Borrie et Magnus S. Magnusson 38 Figure 1: A s c h e ma t ic re p r e s e nta t io n o f a T-p a tt e rn i s s h o w n i n fi g ur e 1 . If o n e assumes that the letters in line 1 correspond to specific performance events (e.g. pass, tackle and shot in football) that appear on the line in proportion to th e time o f th e ir oc c ur re n ce t he n lin e 1 i s a v is ua l r e pr e se n tat io n o f t h e temporal structure of a sports performance. W it h in t h e u p p er l in e t h e re a r e fo u r e v e n ts (a , b , c , d ) th a t o c cu r in a r e g u l a r t e m p o r a l pa t t e r n h o w e v e r t h e p a t te rn h a s b e e n m a s k e d b y th e surro undin g, more ran dom, o ccurre nce of the eve nts w an d k . If a pe rfor- mance analyst or coach were simply visually inspecting the data string it is unlikely that the pattern would have been detected. The T-pattern analysis w o u l d h a v e i d e n t i f i e d t h e p a t t e r n b e c a u s e o f i ts c o n s i s t e n t t e m p o r a l structure. The T-pattern detection algorithms allow an analyst to separate out rando mly occu rring eve nts from temporal pa tte rn s e ven whe n th e ran dom events occur in between elements of the pattern. A T-pattern is essentially a combination of events where the events occur in the same order with the consecutive time distances between consecutive pattern components remaining relatively invariant with respect to an expec- tation assuming, as a null hypothesis, that each component is independently and randomly distributed over time. As stated by Magnusson ’that is, if A is an earlier and B a later component of the same recurring T-pattern then after an occurrence of A at t, there is an interval that tends to contain at least one occurrence of B more often than would be expected by chance’ (Magnusson, 20 00 , p. 9 4). The t empo ra l r ela tion sh ip b etwee n A an d B is d ef ine d as a critical interval an d this con cept lies at th e cen tre of the pattern d etectio n algorithms. Th e pattern detection a lg orithms c an a nalyze both o rd in al and temporal data however, for the algorithms to generate the most meaningful analyses the raw data must be time coded i.e. an event must be coded according to time of occurre nce as well as event type . The coding of many event-typ es and corres pond in g times re sults in the type of d ata set shown in figure 2. This figure displays a behaviour record from the second half of a club football match and consists of 250 series of occurrence times (one for each coded event type) ordered according to their first occurrence time. Schematic representation of a T-pattern viewed within a normal data string and as it appears in isolation.
Detection of real-time patterns in sports interactions in football 39 Figure 2: Only limited aspects of T-pattern detection has been presented here to give some in sigh t in to th e theo retic al bas e of the p roce ss. A c omplete e xp la- natio n of the the ore tical roots of the patte rn -de tec tion a lgo rithms to gether wi t h a n o v e r v i e w o f t h e w i d e r u s e o f t h e p r o c e s s h a s b e e n p r e s e n t e d elsewhere (Magnusson, 1996, 2000). Method A team sport, football, has been analyzed with the intention of identifying whe ther T-pa tte rn d etection has relevan ce as a n a nalytical metho d with in pe rfo rma nc e an a lys is. The res ea rc h u tilize d mu ltip le g ame a n aly sis with each game being treated as a single case. Twenty football matches, thirteen club and seven international matches, were coded using a combination of the f o o t b a l l m a t c h a n a l y s i s s y s t e m d e v e l o p e d a t L i v e r p o o l J o h n M o o r e s University and ThemeCoder, enabling detailed coding of digitized video files (2 5 fra me s pe r s ec .). C o din g in clu d ed da ta on pi tch po sit ion , pl ay er a nd match events. Pitch position was classified according to the pitch divisions sho wn in fig ure 3 . The p rima ry ev ent ca teg ories fo r d ata colle ction were : pa s s ; ta c k le ; h e a d e r; ru n ; d ri bb l e ; c le a r a nc e ; sh o t ; c r o ss ; se t -p l ay ; lo s t control; foul. Additional qualifying statements could be tagged to each event category. All data was analyzed using the Theme software package. Figure 3: A time series behaviour record from the second half of a football match from the European Champions League 1997. The match was coded from a digitized video recording of approximately 45 minutes duration (time is in seconds).
A schematic representation of the zones identified for analysis of ball movement within football. Présentation schématique des zones pour l’analyse des déplacements du ballon. Direction of play Gudberg K. Jonsson, Sigridur H. Bjarkadottir, Baldvin Gislason, Andrew Borrie et Magnus S. Magnusson 40 Results and discussion Th e data show that a high nu mber of tempo ra l patterns e xist in foo tb all. The numb er, freque ncy and comple xity o f th e detected patterns, indicates that sport behaviour is very structured. This synchrony was found to exist on different levels, with highly complex time structures that extended over consi- derable t ime spans wi th in per form a nc es wi th pa t ter ns occurr in g in bot h cyclical and acyclical fashion. The data take n from fo otba ll show th ree discrete ex amp les th at ide ntify within-team patterns (e.g. figure 4, ball movement) and interactive patterns involving both teams (figures 5 and 6, goal scoring).
A temporal pattern relating to attacking movement of the ball through the centre of the pitch.
analysis software showing temporal and hierarchical representation of a T-pattern. The three boxes in this figure involve the same observation period and the T-pattern relates to ball movement in the centre of the pitch within in a football match.
representation of the same data. 1. Player A receives the ball in Zone 8, passes the ball to a team mate and runs forward. 2. Player A receives the ball in Zone 11, passes the ball to a team mate and runs forward. 3. Player A receives the ball in Zone 14, passes the ball to a team mate in Zone 15 (4).
Detection of real-time patterns in sports interactions in football 41 A typical within-team event pattern from the football analysis is shown in fig ures 4 a a nd 4 b. Th is figure displays a d etected T-patte rn that occurred three times during the first half of a European Championship qualifying match (1998). The three boxes in figure 4a show the same observation period. The upper-left box shows the hierarchical con structio n of the pattern. Th e tree stru ctu re identifies th e simp le pa tterns on it’s right ha nd ed ge an d, as the tree builds towards the left edge of the box, shows how the simple patterns are linked together to form the more complex pattern. The up per-righ t b ox displays the time point of ea ch e vent-type in th e pattern and the ir pa tte rn conne ction base d on the critical interval relationship betwee n their oc cur- rence series. The bottom box shows the pattern, as a hierarchical structure, expressed in relation to the observation period i.e. when it occurred during the match (only complete patterns are shown in this box). The pattern describes how player A moves the ball towards the opponents goal by receiving the ball in, and then passing it out of, pitch zones 8, 11 and then 14 consecutively. Player A then completes the sequence by passing it on to player B who receives it in zone 15. The pattern describes an attacking mov emen t th ro ugh the midd le of the p itch, which op ponents wo uld clearly w i s h t o p r e v e n t . Tr a d i t i o n a l f r e q u e n c y a n a l y s i s o f p a s s i n g w o u l d h a v e identified the ball reception and subsequent pass from each zone as discrete events but would not have linked the consecutive actions in the four zones. The movement from zone 11 to 14 also occurred on another five occasions during the first half (figure 4a, upper right box) further suggesting that player A wa s w o r ki n g eff ec t iv el y th ro u g h th e ce n t ra l c h a n ne l of th e p it ch . Th i s integrated form of analysis would potentially enhance the information given to the coach.
Figures 5 and 6 show examples of detected T-patterns from European Cup match b etwee n PSV Eindh ov en an d Ba rce lon a, wh ich finish ed with a 2-2 draw. An analysis of the match revealed patterns including all four goals that were scored, Barcelona goals in figure 5 and PSV goals in figure 6. The data shows the capacity of the analysis process to identify longer, more complex 1. BCL - Rivaldo passes the ball; 2. PSV - Faber intercepts the ball; 3. BCL - Sergi passes the ball; 4. PSV - Jonk makes a bad pass; 5. BCL - Dugarry successful individual act; 6. BCL - Dugarry attempts a shoot, saved; 7. BCL - Sergi intercepts the ball; 8. PSV - Jonk takes a free-kick; 9. BCL - Luis Enrique scores a goal (two occurrences, p<.005).
Gudberg K. Jonsson, Sigridur H. Bjarkadottir, Baldvin Gislason, Andrew Borrie et Magnus S. Magnusson 42 patterns that involve extensive interactions between opposing teams. There is a cl e ar c on s is te n t t e mp or a l pa t te rn p re c e d in g al l g o a ls sc o re d in t hi s match. The patterns cover an extended period of time within which the two tea ms e xch ang e po sse ssio n on at lea st two oc cas ions . The len gth o f th e time per io ds b etwee n the e ve nts f orming th e patte rn wa s su ch tha t othe r match e ve nts will hav e oc curre d be twe en pa ttern e ve nts . The le ng th an d na ture of the pa ttern s is s uch th at one c ou ld q ue stion whe the r, g iv en th e ex tend ed time p eriod o f th e patte rns a nd th e cha ng es of p oss ess ion, th e pa tt e r n s e v e n ts we r e c a u s a l ly re l a t e d . H o w e v e r th e c o n s i s t e n c y in th e temporal pattern preced ing each goal is quite clear. In this case the infor- m a t i o n p r o v i d e d b y t h e T- p a t t e r n a n a l y s i s r a i s e s q u e s t i o n s a b o u t t h e relationship between events that are spread over an extended time period. This doesn’t provide a coach with clear and simple answers about the nature of the goals scored but without the analysis the potential significance of the relationships between events would not have been considered. This analysis may therefore prompt a coach to revisit the video footage of passages of play to iden tify cau sally link ed ele me nts of perfo rman ce tha t migh t h ave b ee n missed had the temporal pattern not been identified.
In addition to immediate analysis of individual matches the data were also used to look at two additional issues relating to structure within team perfor- mance. The first issue investigated related to the potential interrelationship between performance rating by coaches and the degree of structure in team p e r f o r m a n c e . T h r e e e x p e r i e n c e d f o o t b a l l c o a c h e s a n d f i v e a m a t e u r s observed several club and international matches and were asked to rate the performance of every player (on both teams) on a simple ten point Likert type scale. For each coach the player ratings for a specific team were averaged to give a team performance rating. Team performance ratings were then corre- lated (Pearson product-moment correlation) against the number of patterns exhibited by each team. The data (cf. figure 7) show that the coaches’ ratings of team performance were significantly correlated to the number of patterns id e n ti f ie d fo r e a c h te a m ( r =0 . 8 1 , p <0 . 0 5 ) . L o wer co r r e la t i o n was foun d b e t w e e n t h e a ma te u r s r a t in g s o f t e a m p e r fo rm a n c e a n d t h e n u mb e r o f patterns identified for each team (r=0.53, p<0.05). Pattern 3: 1. PSV - Petrovic passes the ball; 2. BCL - Hesp, keeper throws in the ball; 3. PSV - Vampeta passes the ball; 4. BCL - Luis Enrique makes a bad pass; 5. BCL - Dugarry attempts a shoot, saved; 6. PSV - scores a goal (two occurrences, p<.005).
Detection of real-time patterns in sports interactions in football 43 Th e link between pe rfo rma nce rating and pattern pa rticipation sug gests that coaches were recognizing, albeit at a potentially subconscious level, the structure within a team’s pla y. However the traditional rationale for perfor- ma n c e a n a l ys is i s th a t c o a c h es c a n n o t o b s e rv e a n d re me mb e r d is c re te events within critical event sequences (Franks and Miller, 1986). Yet, in this sample, the fact that coach performance ratings were correlated with pattern participation suggests that co aches were perceiving information abo ut the interrelationships between events. This finding also warrants further investi- gation since it relates to such a fundamental foundation in the performance analysis literature.
T h e s e c o n d i s s u e c o n s id e r e d w a s t h e c o mp a r a ti v e l e v e l o f t e mp o r a l stru ctu re with in clu b an d in tern atio na l foo tba ll ma tch es . I n a simp le da ta manipulation three randomly selected club and three international matches w e r e c o mp a re d in t e r ms o f th e me a n n u m b e r o f p a t te r n s a n d t h e me a n n u mb e r o f p a tt e r n o c c u r re n c e s id e n t if i e d in e a c h ma tc h t y p e . T h e d a ta (cf. figure 8) show that in ternational fo otball ha s a more d efin ed te mp oral st ru c tu r e t h an c lu b fo o tb a ll . Th is fin d in g ma y b e d u e to th e p re s e n ce o f higher technical abilities in international footballers which help create a more structured game or, alternatively, contextual differen ces between club and international football e.g. club football is played at a higher pace throughout mitigating against the development of structure within the game. Whatever the reason the clear difference in temporal structure between club and inter- national football merits further investigation. Figure 8: 0 0,2 0,4 0,6
0,8 1 Coaches Novice Ratings C o rrel at io n w it h P at tern Participat ion Correlation between team pattern participation and coaches and novice subjective ratings (p>0.05).
0 1000 2000 3000
4000 5000
6000 Events
N. dif. Pat. Pattern occ. Num b e r National Matches Club Matches Number of events, different pattern types and pattern occurrences between national team matches and club matches. All differences are significant at p>0.001. Nombre d’évènements, nombre de types de patrons différents et nombre d’occurrences de patrons dans des matchs de club et nationaux (p > 0,001). Gudberg K. Jonsson, Sigridur H. Bjarkadottir, Baldvin Gislason, Andrew Borrie et Magnus S. Magnusson 44 Discussion The number, frequency and complexity of detected patterns indicates that the behaviour of football players is more synchronized than the human eye can detect. This synchrony was found to exist on different levels, with highly complex time structures that extended over considerable time spans, often in a cyclical fashion, as well as less complex patterns with a shorter time span. S yn ch ro n y o f th is k in d w a s fo u nd to co rr ela te hi gh ly wi th a ss e ss me nt o f perfo rma nce. A stron ger correla tio n was also disc ove re d between pe rfor- mance assessments of professional coaches and team pattern participation than between assessments of amateurs and team participation in patterns. National matches were found to be more structured than club matches. The results show that pattern analysis can be used to track elements in the game such as timing of events, the passing of the ball, team structure etc. in a novel way, indicating that pattern analysis is useful in enhancing existing methods used in football analysis. New kinds of profiles, for both individuals and te ams, c an be d is cove red us in g the de tected b eha vioura l patte rns in combination with elementary statistics. Moreover, some answers are already suggested to questions, such as: Are there certain patterns that are related to doing well or bad? What responses seem to be evoked by certain actions or se que nc es of ac tion s? Coac he s cou ld use th is kind o f struc tura l inf or- mation when selecting players or when searching for the opponent’s "weak spots". The preliminary data highlights the potential for T-pattern analysis to make a significant contribution to sport performance a nalysis. Current analytical methods that focus on simple frequency analysis cannot identify the temporal pa tte rn s with in a s p or ts p e rfo rma n c e. Co n se q u en tl y wit ho u t th is fo rm o f analysis meaningful information is not being made available to the coach. If this information is not available then it possible that performance is not being optimized. T h e d a ta a ls o p o in t t o w a rd s th e n e e d t o in v e s t ig a t e th e p o te n t ia l li n k between temporal structure in sport performance and the understanding of performance being generated by coach observations. The data suggest that whilst coaches may not be able to accurately recall discrete events they do pe rce ive inte r-re la tion sh ips be twee n e ve nts. This a na lys is a pp roa ch ca n a s s i s t i n g e n e r a t i n g a g r e a t e r u n d e r s t a n d i n g o f c o a c h k n o w l e d g e construction. References Borrie, A. & Jones, K. (1998). It’s not what you do it’s the way that you do it: is frequency of occurrence and adequate index of performance in observational analysis. Journal of Sports Sciences, 16, 1, 14 Franks, I.M. & Miller, G. (1986). Eyewitness testimony in sport. Journal of Sport Behaviour, 9, 39-45 Jonsson, G.K. (1998). Detecting patterns in complex behavioural processes with the observer and theme. In L.P.J.J. Noldus (Ed.), Measuring Behaviour ´98. 2
Jonsson, G.K., Bjarkadottir, S.H. & Gislason, B. (2000). Detection of real-time interaction patterns in football. In L.P.J.J. Noldus (Ed.), Measuring Behavior 2000, 3
Detection of real-time patterns in sports interactions in football 45 Magnusson, M.S. (1983). THEME and syndrome: two programs for behaviour research. Symposium in Applied Statistics, H.C. Oersted Institut, University of Copenhagen. Magnusson, M.S. (1988). Le temps et les patterns syntaxiques du comportement humain : modèle, méthode et le programme THEME. In Revue des Conditions de Travail. Actes du premier colloque national d’Ergonomie scolaire, Université de Lille, 19-20 mars 1987. Marseille, Octares. 284-314. Magnusson, M.S. (1996). Hidden real-time patterns in intra- and inter-individual behaviour: description and detection. European Journal of Psychological Assessment, 12, 112-123. Magnusson, M.S. (2000). Discovering hidden time patterns in behavior: T-patterns and their detection. Behavior Research Methods, Instruments & Computers, 32, 93-110. 46 Download 179.11 Kb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling