=Paper=
{{Paper
|id=Vol-2841/BigVis_6
|storemode=property
|title=Visual Analysis of Player Interactions in Soccer Games
|pdfUrl=https://ceur-ws.org/Vol-2841/BigVis_6.pdf
|volume=Vol-2841
|authors=Ivona Ivkovic-Kihic,Daniel Seebacher,Manuel Stein,Tobias Schreck,Reinhold Preiner
|dblpUrl=https://dblp.org/rec/conf/edbt/Ivkovic-KihicSS21
}}
==Visual Analysis of Player Interactions in Soccer Games==
https://ceur-ws.org/Vol-2841/BigVis_6.pdf
Visual Analysis of Player Interactions in Soccer Games
Ivona Ivkovic-Kihic Daniel Seebacher Manuel Stein
Graz University of Technology University of Konstanz Subsequent GmbH
University of Sarajevo daniel.seebacher@uni-konstanz.de Konstanz, Germany
ivona.ivkovic-kihic@cgv.tugraz.at manuel.stein@subsequent.ai
Tobias Schreck Reinhold Preiner
Graz University of Technology Graz University of Technology
tobias.schreck@cgv.tugraz.at r.preiner@cgv.tugraz.at
Figure 1: We develop an interactive visual analysis approach for interaction events in soccer data, which are extracted from
player tracking data based on a proximity condition. An interactive framework supports explorative visual analysis based
on categorical (a), spatiotemporal (b), feature-space (d), and detailed animation views (e). Based on appropriately designed
visual (c) and quantitative (d) encodings of player interactions, our approach establishes suitable analysis workflows for
investigating key-interactions that lead to successful outcomes – in this figure, a shot on goal (e).
ABSTRACT 1 INTRODUCTION
Recently, visualization of sport data in general, and soccer data in In team sports such as soccer, a fundamental task of coaches is
particular, has received much research attention. Visual sport data the analysis of individual games to identify strengths and weak-
exploration helps to understand behavior and performance of ath- nesses, plan team lineups and tactics, and understand the critical
letes and teams, identify possible influence factors, and changes key events that lead to success or failure. Besides analyzing the
over time, among other important tasks. In soccer match data, overall behaviour and interplay of an entire team, coaches are of-
much of the play is determined by direct interactions between ten interested in observing local interactions between individual
players spatially close to each other, competing for influence. We players of competing teams, such as tackles for the ball, as the
introduce a novel visual analytics system for exploring pairwise success or failure of these interactions can have a very high im-
player interactions using a trajectory-based data representation in pact on the match outcome. The important key actions deciding
a highly interactive multiple view approach. Our notion of player the outcome of such interactions often happen within very few
interaction is based on proximity of pairs of players, and respec- seconds, where the skill and training of a player can be decisive.
tive motion patterns represented as trajectories. Our approach Accordingly, coaches want their teams’ tactic to be organized in
segments player interactions from soccer match data, as the basis a way that the strengths and weaknesses of their players regard-
for linked analytical views. A matrix view allows to explore in- ing different interaction types perfectly outbalance the opposing
teraction frequencies between players, group of player roles, and team. Besides pure match analysis, a detailed assessment of a
assess overall game dominance between teams. An appropriately player’s interaction behaviour is also important for scouting for
defined interaction glyph allows to compare interactions based new players that optimally complement their team.
on player motion, ball possession, and pitch position. We further Several previous work have investigated visual analysis ap-
investigate the design of a descriptor encoding the geometric proaches for interactions either based on raw motion data [21,
configuration of interaction trajectory pairs, enabling common 24], employing an interaction definition merely based on imi-
analytical tasks like clustering or searching for similar interac- tative motion [13], or focused on semantic aspects at the level
tions. We demonstrate the applicability of our approach by use of global team tactics [31]. However, especially within invasive
cases on real soccer match data, detailing the analytical tasks team sports, considering short-time small-scale interactions and
supported by our system. their relations is highly relevant for the analysis process. Analyz-
ing these interactions comes with different challenges that stem
from the complexity and interdependencies of their contained
information. Simple pattern detection on the motion data is often
© 2021 Copyright for this paper by its author(s). Published in the Workshop Proceed-
ings of the EDBT/ICDT 2021 Joint Conference (March 23–26, 2021, Nicosia, Cyprus)
not feasible for this analysis task, as the crucial motion informa-
on CEUR-WS.org. Use permitted under Creative Commons License Attribution 4.0 tion decisive for the interaction outcome is mostly concentrated
International (CC BY 4.0)
in a very short time span, and is further dependent on additional 2.2 Visual Analysis with Glyphs
semantic context such as the change of ball possession. Glyph-based techniques are a well-known approach in visual-
In this paper, we develop a visual analysis approach that ad- ization, often designed to give compact overviews over large
dresses these challenges and aims at fostering an interaction- amounts of data records and/or dimensions. According to [4],
centric analysis process, allowing for investigating game-specific, "Glyphs are a common form of visual design where a data set is
player-specific and shape-based relations of individual interac- depicted by a collection of visual objects referred to as glyphs".
tions in a soccer game. To provide an accessible notion of the Different visual channels are typically used to compose glyphs,
nature of individual interactions, we propose a suitable visual e.g., color, shape, size/height/length, orientation, texture, opacity
representation as glyphs, encoding both motion data and seman- etc. Symbolic glyphs can also be used to represent trajectories
tic context information (Section 3.2). We further investigate an and movement [7]. Recently, Motion Glyphs [6] were introduced
appropriate quantitative encoding of its motion data, providing to show properties of large dynamic graphs. The glyph design
the analyst with a feature-related structuring of the data and giv- includes an outer circle showing context of the graph, and a focal
ing rise to common analysis tasks like similarity assessment or part of the graph as a node-link diagram in the center. In a case
clustering (Section 3.3). To foster an in-depth analysis of matches study, it was applied to sets of moving elements (fish schools),
and player performances based on these interactions, a visual supporting analysis of leader/follower patterns among others.
representation of a player-related spatiotemporal context of these The Motion Rugs approach [5] is a dense visualization which
Interactions Glyphs is proposed, summarizing the interaction his- provides a space-efficient overview of development of moving
tory of individual or pairs of players (Section 4). Based on these elements over time, supporting analysis of patterns in groups of
atomic encodings, the interaction data is made accessible in an movers.
interactive analysis framework, combining means of navigation In our work, we rely on glyphs to show trajectory interactions,
and filtering in a categorical, spatial, temporal and feature-space using color, shape, orientation and size, as well as the outcome
domain (Section 5). We demonstrate that our approach allows of an interaction in terms of change in ball possession.
for quick insights into the current game situation and its rela-
tions to the performances of players within individual or groups
of interactions (Section 6). In particular, our system supports a
2.3 Soccer Analytics
causal investigation of game outcomes, aiming for an efficient The analysis of sports data in general [9, 18], and soccer data in
identification of key interaction events that led to a successful or particular [17], has become an important application in visual
unsuccessful game outcome. As a result, our approach allows for data analysis. Soccer Stories [17] represents one of the first visual
a deeper understanding of games in team sports, and gives rise to soccer analysis systems, giving visual designs for different soccer
new workflows for sports-related analysis and decision-making. match situations. Interaction allows to explore soccer matches
by phases and events, e.g., corner kicks, passes, dribbling etc. A
2 RELATED WORK large amount of work in Soccer Analytics focuses on analyzing
team tactics and the global behaviour of a team [14, 16, 20, 22, 29].
Our work relates to several topics, including spatiotemporal vi- Marcelino et al. [15] analyze behavior patterns of football players,
sual data analysis, glyph techniques, and applications in soccer measuring performance fingerprints of individuals and teams,
data analysis. We next discuss selected related works and how and considering pairwise interactions to model and assess overall
we add to it. team performance. Similar in spirit, our work focuses on inter-
action pairs as the basis on which team analysis builds. In our
2.1 Visual Analysis of Spatiotemporal Data approach we support exploration of interaction pairs by inter-
Geospatial data arises in many areas, and to date, visual analysis active cluster analysis and linked views for detail exploration.
of this data has received much attention. There is already a rich In the literature, to date many player and match features are
body of work on visual analysis of geospatial data in general [3], considered for visual exploration, including free and interaction
and movement data in particular [1]. Key analysis tasks in visual spaces [27], pressure [2], collective team movement [26], perfor-
movement data include at which level of detail to describe move- mance and event data [11], and much more. While many works
ment, how to compare movements, and identify similarities and consider abstract pitch and trajectory visualization, some works
outliers, both for trajectories in isolation, or groups of trajectories. map derived data and visualizations onto soccer video streams,
To date, many applications have been studied, e.g., exploration of for integrated analysis. In [28], such a mapping is proposed, and
dynamics of traffic flows [10, 23]. Also, in [25], animal movement shown that coupling video with visualization overlay allows for
patterns are considered. Often, functional relationships need to effective match context in the analysis.
be considered for object movements, which may be influenced For the analysis of individual player movement, an important
by varying environmental influences on the movement. aspect is to provide a proper visual representation and abstraction
Besides movement in physical space, movement can also be of player’s trajectories [21], which also gives rise to designing
an important factor when working with time-dependent visu- suitable approaches for interactive search within the trajectory
alizations. In [30], patterns in time-dependent scatter plot data data [24]. We resort to similar abstractions for putting key in-
were identified by movement analysis, e.g., allowing analysts to teraction events between players into a spatiotemporal order.
group similar changes and segment meaningful time intervals of Other previous work has investigated the classification of partic-
the change. ular match events like passes in football matches based on given
In this work, we investigate a specific feature of group move- spatiotemporal data [8].
ment data, that is, the motion of two locally interacting entities. In our work, interactions between players and their outcome
We analyze the interactions of such entities in terms of the geo- are utilized as the key aspects of the analysis of a soccer game.
metric configuration of their trajectory intervals at the time of We propose an appropriate design to represent these interactions
their encounter. as glyphs, which can be displayed in their spatial context on the
soccer pitch. We are investigating a feature encoding for the tra-
jectory footprints of two players during a mutual interaction, en-
abling tasks like similarity-based exploration. A proposed visual
analysis system is complemented by a matrix view structuring
the player interactions based on player roles, and providing an en-
trance point for an interactive exploration of player interactions
within a game.
3 INTERACTIONS IN SOCCER GAMES
Figure 2: Interaction glyph design
3.1 Definition and Input Data
In the context of our work, interactions are defined based on a
set of trajectories that track the motion of players over a certain when being passed or shot. In order to reduce overplotting in
interval of time, in our case, the time frame of a soccer match. such cases, the ball trajectory is clipped whenever it exceeds the
In this paper, we are using data extracted from a vision-based glyph circle, and a non-transparent arrow indicates the direction
motion tracking technique. The data contains the position of of ball movement outside the ring.
the soccer players as well as the ball with a spatial resolution of To visualize ball possession, the outer ring is divided into two
10 𝑐𝑚 and a temporal resolution of 100 𝑚𝑠. Moreover, the data is parts: the first, smaller section (Fig. 2a) colored corresponding to
annotated to indicate for each point in time the player that holds the team holding the ball before the interaction, and the larger
the ball. An interaction is defined to occur whenever the distance section (Fig. 2b) of the ring colored based on the team holding the
between two players falls below a certain proximity threshold. ball right after the interaction. In this way, it can be easily deter-
Starting from the point of closest distance at a reference time mined whether the interaction caused a change in ball possession.
𝑡 0 , we extract the trajectory segments of the interacting players Similarly, a uniformly colored ring indicates the continuous ball
from the time interval [𝑡 0 − Δ𝑡, 𝑡 0 + Δ𝑡] and use these segments possession of the respective team. Besides the local inspection
for visually representing the players’ motion at this interaction. of player motion, this kind of glyph design maintains a visual
For the data demonstrated in this paper, we constantly use a overview of ball possession changes when exploring larger sets
Δ𝑡 = 0.75 seconds. To obtain a robust set of non-redundant of glyphs.
interactions, interactions with overlapping time intervals are
removed, keeping the interaction centered around the minimal 3.3 Quantitative Encoding
distance between players. An important aspect for analyzing the nature of interaction
Moreover, we only extract interactions between players of events between players is the ability to detect and assess simi-
opposing teams that include the ball, as these are the most impor- larities between different interactions, and thus allow for clas-
tant ones to affect the current and future game situation, e.g. by sifications of motion patterns whose relation to different game
change of ball possession, or by preparing situations that lead to situations or outcomes of tackles are to be analyzed. To establish
a goal. In this way, less relevant proximity interaction between such a measure of similarity, we seek for an encoding of the
players of the same team (e.g. players in a wall during a free kick common motion patterns of interacting players.
scenario) are not taken into consideration. If the tracking data A typical approach is a quantitative encoding of motion tra-
is labeled accordingly, we also filter out interactions that hap- jectories as shape descriptors that capture the geometric features
pen during non-active phases, e.g. after outs or fouls and during of trajectory segments [19]. However, in contrast to descriptors
player substitutions. Finally, the motion data from the second half for single trajectories, the encoding of pairs of interacting trajec-
of the game is mirrored to allow a simplified spatial mapping of tories in a single descriptor raises additional challenges. Besides
the player motion and their interactions in the further course of rotation and reflection invariance, which is a commonly desir-
their visual analysis. Note that due to reasons of confidentiality, able property for shape-based descriptors, we also require it to
the datasets used in the rest of the paper are anonymized. be player and team agnostic, i.e., exchanging the motion data
between involved players should lead to the same interaction
3.2 Visual Encoding descriptor. To this end, we encode the trajectory pairs of an
In order to represent and analyze a set of player interactions interaction in a feature vector, capturing the most descriptive
throughout the game, a consistent visual representation of both properties of the involved motion data characteristic for a tack-
motion data, as well as its semantic context in the game, is needed. ling event: (1) the relation between the involved players’ motion
To serve that purpose, glyphs combining these two types of data directions (in-sync, intercepting, or frontal approaching), and (2)
are used. Our proposed glyph design consists of (1) an inner part the change in the scope of action for the player owning the ball,
representing the trajectories of two interacting players as well indicating how pressing the attack is. In our design, we measure
as the ball trajectory, and (2) an outer ring that represents the these properties along 8 regular intervals during the interaction
ball possession before and after the interaction phase. Players’ time frame, illustrated in Figure 3b. At each time step, we capture
trajectories (Fig. 2c and Fig. 2d) are plotted based on the fixed the distances 𝑑𝑖 between player positions as well as the current
number of points registered at the same points in time for both motion directions 𝑣®𝑖 , 𝑤® 𝑖 from the 𝑖-th to the next sample point.
players, with an arrowhead pointing at the direction of the play- From these data, the angles 𝛼𝑖 between player orientations as
ers’ motion. Curves representing the movements of players are well as the distance differentials Δ𝑑𝑖 = 𝑑𝑖+1 −𝑑𝑖 are extracted and
colored by the players’ teams. The ball trajectory is presented by renormalized to the unit cube [0, 1] 16 based on [−𝜋, 𝜋] for angles,
a thicker, red semi-transparent line, with the arrowhead showing and on the min/max of all distance differentials in the dataset.
direction of its movement. In general, during the interaction time The final normalized values are then encoded in the interaction
interval, the ball can move larger distances than the players, e.g., descriptor (Δ𝑑 1, 𝛼 1, . . . , Δ𝑑 8, 𝛼 8 ).
5 PLAYER INTERACTION ANALYSIS
wi In this section, we present an interactive system that combines
different views utilizing the visual and quantitative encoding of
di the interaction data (Section 3) as well as their spatiotemporal
di+1
embedding (Section 4) enabling an explorative visual analysis
vi
wi a i workflow. In order to provide the user with a suitable overview on
the data, we require different views on the interactions present
in a game, providing a categorical overview, showing the fre-
quencies of mutual interactions between players of opposing
teams and their success in keeping or stealing the ball, a spa-
(a) (b)
tial overview to show where these interactions happened, and a
temporal structure to indicate when the interactions happened.
Figure 3: (a) Symmetric shape descriptor for interacting
At the same time, a user might want to explore data through
trajectory pairs. (b) Resulting feature space of interactions
multiple levels of details, including: (1) an overall view of the
shown on their dominant eigenplane. The interactions
entire spectrum of interaction data present in a single game, as
exhibit a continuous distribution, smoothly varying be-
well as (2) their spatial distributions and concentrations, (3) the
tween different motion classes shown at the outer hull.
quantitative distribution of interactions over the participating
players and player pairs, (4) groups of interactions based on our
feature-based cluster analysis, as well as (5) detailed analysis of
interactions (details on demand).
Based on the above requirements, we design a user interface
This encoding gives rise to basic analytical tasks like clustering, that provides these views and interactively links them to allow
performing similarity queries based on the motion characteristic, for an encompassing visual analysis and exploration of the data.
and similar. Figure 3b illustrates the interaction data extracted
from a given soccer game on the dominant eigenplane of the re-
sulting 16-dimensional feature space. The data exhibits a mainly 5.1 Interaction Matrix
continuous distribution of interaction data in this space, but also
reveals prototypic, strongly discriminating interaction trajectory The Interaction Matrix (Figure 4b) shows data at a global level
pairs around the convex hull of this subspace, as highlighted and at a high level of abstraction, and is used as a starting point
in the figure. For instance, synced parallel motion (far left) vs. for player-based analysis.
frontal opposing parallel motion (far right), or cross-overs (bot- Cells of the upper triangular matrix represent the overall view
tom left) vs. local dribbling with U-turns (top right). Extracting of the number of interactions of each player and each pair of
such interaction prototypes around this convex hull enables a players, respectively. Color intensity of non-diagonal cells is
rough distance-based clustering (c.f. scatterplot color coding), determined based on the number of interactions that occurred
and will later also be utilized as anchor points for an explorative between the two players, where darker colored cells indicate
visual analysis process. that more interactions happened between the two players. Diag-
onal cells represent the number of interactions that one player
had with all other players during the game, where more intense
4 SPATIOTEMPORAL EMBEDDING color means more interactions were registered, and the color is
Glyphs as described in Section 3.2 represent the two players’ determined by the team color of the player.
movements, ball movement, ball possession, and spatial place- In contrast, the lower triangular matrix represents the players
ment. Glyphs as such still miss the context information like other performances in the means of ball possession changes during the
players’ movements, and the chronological ordering of the events. interactions. Each cell is colored based on the team color of the
However, these pieces of information are crucial for understand- player that had more positive outcomes. In this context, a positive
ing the context of interaction and making meaningful and com- outcome for playerA means that (a) playerA’s team kept the ball,
prehensive conclusions. or (b) playerA’s team stole the ball from playerB’s team. A higher
To this end, we design a suitable visual embedding of these color saturation represents a higher percentage of interactions
glyphs into their spatiotemporal context, i.e., their position on with a positive outcome for the respective team. The matrix
the soccer pitch and the timeline, which at the same time estab- is complemented by bars behind the player’s names indicating
lishes a connection to the involved players. Our design comprises the total ball possession time for each player. Combining this
a collection of all interaction glyphs associated with a player, information with the number of interactions allows for a more
or a specific pair of players, connected by a spline curve that distinctive assessment of the player’s performance. Players in
puts them into chronological order (see Figure 4c). We therefore the Matrix are ordered first by the team, then by their playing
call this visual representation an Interaction History Curve. The positions, and different roles are distinguished by line separators
curve segments are colored by a gradient of yellow and green and labeled accordingly. This particular ordering allows for both
for the first and second half respectively, corresponding to the a player-based performance analysis as well as an overall role-
timeline shown in Figure 4a and thereby establishing a visual based assessment between teams.
representation of time. When hovering over the cells on the upper part of the ma-
In the following, we are proposing an interactive approach trix, corresponding cells in the lover part are highlighted, and
that integrates the visual representations developed so far into vice versa. In this way, users can easily explore both interaction
an visual analysis tool allowing for a task-oriented exploration numbers and possession ratio for the selected pair of players at
and analysis of soccer interaction data. the same time. Finally, selecting a non-zero cell in the matrix, all
Figure 4: Overview of our visual analysis interface for soccer player interactions for a specific game and selected pair of
players. (a) Timeline and Toolbox. (b) Interaction Matrix. (c) Soccer Field View. (d) Interaction Prototypes. (e) Selection
grid. (f) Similarity Search Grid (g) Interactive scatter plot. (h) Zoom panel.
interactions of the corresponding player or pair of players are 5.4 Feature Space and Interaction Prototypes
shown on the Soccer Field View and Similarity Search Panel. Feature Space View. Finally, we add another view on the data,
by structuring them in the visualization of their feature space
as established in Section 3.3. The view consists of two parts: an
5.2 Soccer Field View interactive scatter plot (Figure 4g) showing the feature space
This view (Figure 4c) shows interaction glyphs as described in of the data, and the zoom panel showing selected interaction
Section 3.2 plotted at the spatial location they occurred on the while the user hovers over the scatter plot (Figure 4h). This panel
pitch. Glyphs also serve as trigger points for the situation anima- should provide a more elaborate view on the data records.
tion. When clicked, they show an animation of movements of all
Interaction Prototypes. This panel, shown in Figure 4d, lists the
players and the ball at the time of the interaction, providing the
cluster prototypes extracted around the convex hull as described
user with a detailed visualization of the game situation. More-
in Section 3.3. These prototypical interactions can serve as a
over, a marker on the timeline appears to clearly indicate when
starting point for clusters exploration. By selecting a cluster
the interaction takes place in the game. When activating the
prototype, the cluster members are being plotted on the Soccer
corresponding setting in the Toolbox, Interaction History Curves
Field View as well as the interaction selection grid to allow for
are displayed connecting all the interactions in chronological
further visual analysis. As a result, users can search for patterns
order. Glyph rings representing the ball possession are oriented
in similar interactions from one cluster.
according to the tangent of the Interaction History Curve at that
point.
5.5 Timeline and Toolbox
The Timeline on top of the Soccer Field View represents the
5.3 Similarity Search Panel time of individual interactions and it is colored with the gradient
of yellow and green, similar to the Interaction History Curve.
Selection Grid and Similarity Search Grid. A selection grid,
The Timeline can be used to filter the time span of interactions
shown in Figure 4e, represents the same set of interactions as
shown on the Soccer Field View, which is also useful for reducing
the Soccer Field View, ordered by their time stamp. They serve
the plot density on this spatial view. On top of the Timeline,
as trigger points for a query search that finds the most similar
goal indicators are shown as clickable markers that trigger an
interactions according to the descriptor introduced in Section 3.3.
animation replaying the last few seconds before the respected
Similarity is measured by the Euclidean distance between the
goal. These can be used for detail inspection of interactions that
interaction descriptor. After selecting a query interaction, the
lead to a goal.
closest interactions are plotted in the Similarity Search Grid
Finally, a toolbox, shown in Figure 4a is a simple set of filters
(shown in Figure 4f). Using the query search, a user can explore
and user settings made for the purpose of easier analysis of dif-
similar interactions to the interesting one and search for the
ferent situations. The Interaction History Curve can be switched
common behaviors in the game.
off to reduce the visual load on demand. Outcomes can be filtered
in order to show only interactions that led to a change or no
change in ball possession This is used to analyze the success rate
of players of teams in general.
5.6 Implementation
We implemented our system as a web-based JavaScript appli-
cation using D3 for the interactive matrix plot, the interaction
curves, and other responsive elements. Interaction information
is extracted from tracked soccer game motion datasets in an
offline preprocess based on a predefined proximity range. For
each interaction between two players from opposing teams that (a) Match 1 (b) Match 2
involves the ball, we store the player’s trajectory segments for a
fixed time interval around the point of closest proximity. Based Figure 6: Comparing the player interactions between the
on these trajectory pairs, we then compute the corresponding first and second match shows a clear increase in the over-
interaction descriptors and extract interaction clusters from the all frequency of interactions, as well as a shift away from
resulting feature points as described in Section 3.3. The resulting the anti-diagonal.
trajectory pairs, interaction descriptors, and cluster assignments
are then loaded from the precomputed files into the framework
for interactive analysis. example is given in Figure 6, which compares the first match
and the return match between the same teams. At first glance, it
6 USE CASES AND RESULTS is noticeable that the number of interactions between the first
and second game differs significantly. However, the particular
In order to demonstrate the usefulness of our approach, we iden-
matrix ordering also immediately reveals for which combinations
tify and discuss several typical analysis use cases. For this ex-
of player roles these interactions have increased in particular,
ample, we analyzed an anonymized match from a well-known
which allows to draw direct conclusions about the game. In a
European club competition.
balanced game, the majority of interactions would be expected
Assessing Game Dominance. An interesting finding we imme- to occur along the anti-diagonal of the player-role matrix (high-
diately discovered with our approach during the analysis of this lighted in purple), where defenders face forwards and midfielders
match was the spatio-temporal distribution of interaction glyphs encounter midfielders. In contrast, the second game also shows
depicting transition phases in the first and second half. We com- a noticeable increase in the frequency of interactions between
pared the distribution of the glyphs in the first half of the match midfielders and forwards of both teams (green). Overall, these
(see Figure 5a) and the second half (see Figure 5b), using the are indicators that the second game was more aggressive since
time range filter in the timeline interface. In the first half, a rela- not only the frequency of interactions increased, but the player
tively even distribution of transition interacting glyphs can be combinations within which they happened also shifted away
observed. However, in the second half of the match, the majority from the usual anti-diagonal. This might be explained by the fact
of possession changes occurred in the orange team’s half of the that one of the teams would have went out of the competition if
pitch. This difference in the distribution of the glyphs shows that they would have lost this return match in the group phase.
the blue team was way more dominant during the second half.
Player-Centric Assessment. Another important analysis task
This assumption is also reflected in the outcome of the match.
is the investigation of individual player performances. For the
After a 1:2 in the first half, the blue team was able to score two
game data investigated in this paper, the matrix plot indicates
additional goals in the seconds half, earning them a 3:2 score.
a particularly strong involvement of the orange Wide Forward
Left with the Blue Wide Midfield Right player (Figure 7). A look
on their common interaction history curve provides interesting
insights into the interaction history between these two players.
First, most interactions happened on the side of the orange team,
(a) Half 1 (b) Half 2
Figure 5: The contrast in the distribution of changes in the
ball possession clearly shows that the blue team was more
dominant in the second half of the game.
Role-based Assessment. The matrix-based visualization of in-
teractions provides an overview of the respective interactions Figure 7: Interaction History curve between the orange
of all players in a game. The ordering of the columns and rows Wide Forward Left and the blue Wide Midfield Right
by team and player roles allows the analysis of substructures player over the course of the game, showing a dominance
in the player interactions, which would be difficult or impossi- of the blue midfield player and a large number of unsuc-
ble to inspect with classical matrix reordering techniques. An cessful tackle attempts of the orange player.
where the orange forward player was forced into a typical de-
fender role far back in his own side of the field. Here he attempted
several tackles on the pressing blue midfield player. However, as
the ring colors of the interaction glyphs indicate, none of these
attempts led to a successful gain of ball possession. In contrast,
(a) Parallel Runs
in a single interaction between these players, the blue midfield
player even obtained ball possession from the orange forward
player, in a relatively close distance to the orange goal (see black
arrow). This reflects the above insights into the the game domi-
nance in the second half of the game, on a player-focused level
of detail, showing that the orange forward player was heavily
(b) Cross-Overs
outgunned by the blue midfield player.
Identifying Key Interactions. A further interesting situation Figure 9: Similarity queries on selected glyphs yield in-
we discovered during the analysis of the interaction glyphs is teractions with most similar trajectory configurations, en-
depicted in Figure 8. The glyph shows that the orange team lost abling further exploration of similar interactions.
possession of the ball in the midfield. While investigating the
replay animation of this scene, we noticed that the orange player results exhibit a similar cross-over shape. Comparing the sets of
shown in the glyph first receives a pass from a player of his parallel runs and cross-over glyphs, we can also observe overall
own team. However, the blue player in the glyph is immediately longer trajectory path lengths in the parallel runs. As trajectory
putting pressure on the orange player. The orange player’s sub- segments correspond to constant time intervals (1.5 seconds in
sequent pass was a miss pass, which may have been caused by our examples), this indicates that these type of interactions are
the applied pressure. After this miss pass, the blue team was able generally much more fast-paced than the cross-overs.
to bring the ball directly to the strikers, which created a very
dangerous situation with two blue players in ball possession and 7 DISCUSSION AND FUTURE WORK
a free path towards the opposing goal.
In the previous section, we have shown a set of basic but impor-
tant analysis use cases relevant to soccer coaches and analysts.
These range from an overall assessment of a team’s performance,
to investigating the role, importance and performance of individ-
ual players, up to detailed analyzes of key interactions and their
influence on the game outcome. Our system provides several
different views on the interaction data present in a game, and
allows for temporal, player-related or shape-based filtering. The
latter is enabled by defining appropriate shape descriptors for
pairs of trajectory segments that constitute an interaction event.
Limitations. Targeting at the visual analysis of potentially
complex player interactions over a whole game, our approach
Figure 8: Loss of ball possession by the orange team results currently exhibits several limitations. One issue is the problem of
in a dangerous game situation, as there are now two blue visual clutter when many interaction glyphs are superimposed
players in ball possession with a free path to the goal. in the spatial embedding, as seen in Fig. 7. These need to be ad-
dressed using appropriate means of visual reduction, like density
based scaling or visual simplification, and combined with suitable
Shape-based Interaction Exploration. In many cases, the type
interactive detail-on-demand techniques.
and interpretation of an interaction between players is directly
Moreover, our current way of extracting semantic interactions
reflected by their trajectory segments shown in the glyph. Based
from the raw trajectory data enforces the assumption that any
on a selected interaction glyph of interest, our system allows
interaction only involves two players. However, in general drib-
the user to investigate additional interactions of the same type.
bling events or close-range interactions inside the penalty box,
Selecting a glyph in the Selection Grid (Fig. 4e) issues a search
e.g. after a corner kick, more than two players can be close to
for similar interactions throughout the game, utilizing the inter-
or in physical contact to the ball. Our current approach would
action shape descriptor developed in Section 3.3. The 20 most
break these down to a set of two-player interactions, which is
similar interactions are then presented in the Similarity Search
not able to visually express the complexity of the interaction in
Grid (Fig. 4f). These can then be further investigated by clicking
a single glyph, or encode it in a descriptor.
them, inspecting their geographic context on the soccer pitch
Another aspect is the usability of abstract views on the data
(Figure 4c), and inspecting them in detail using our system’s
provided by our system. While shape-based similarity search and
animation capabilities. Figure 9 shows the retrieval results of the
data- or feature-based views are typical elements used by visual
for two different interaction queries. In Fig. 9a, the user searches
analysis experts, they might be not as intuitive to domain users
for glyphs similar to a parallel run of two players. The result set
like soccer coaches. However, work from other domains indicate
shows 20 configurations of high geometric similarity, and proofs
that such elements can indeed be useful for domain experts [12].
the rotation invariance of the proposed descriptor. Moreover, the
result set indicates that interactions of these type rarely lead to a Future Directions. Other interesting future work directions in-
change in ball possession. The query in Fig. 9b denotes an interac- volves finding a taxonomy of interaction patterns, e.g. by enhanc-
tion of players with crossing trajectories. Again, most similarity ing the cluster analysis of interactions by visual cluster analysis
tools. For instance, allowing users to label interaction examples [11] Halldór Janetzko, Dominik Sacha, Manuel Stein, Tobias Schreck, Daniel A.
from the PCA view and using this data to train an interaction Keim, and Oliver Deussen. 2014. Feature-driven visual analytics of soccer
data. In 2014 IEEE Conference on Visual Analytics Science and Technology, VAST
classifier would further improve its usability. 2014, Paris, France, October 25-31, 2014. 13–22. https://doi.org/10.1109/VAST.
Conceptually, besides single matches, we may also look at 2014.7042477
[12] Wolfgang Jentner, Dominik Sacha, Florian Stoffel, Geoffrey Ellis, Leishi Zhang,
the differences in interaction patterns between matches of the and Daniel A. Keim. 2018. Making Machine Intelligence Less Scary for Crim-
same teams, and compare them with coaching strategies. An inal Analysts: Reflections on Designing a Visual Comparative Case Anal-
obvious extension of the player interaction matrix is to code it ysis Tool. The Visual Computer Journal (2018). https://doi.org/10.1007/
s00371-018-1483-0
for types of interaction patterns, e.g., to observe if same players [13] Hoang M Le, Peter Carr, Yisong Yue, and Patrick Lucey. 2017. Data-driven
more frequently interact in similar patterns, or whether their ghosting using deep imitation learning. In MIT Sloan Sports Analytics Confer-
patterns change or evolve over time. We presume that sequence ence. https://resolver.caltech.edu/CaltechAUTHORS:20170316-121646643
[14] Jose Luis Sotomayor Malqui, Noemí Maritza Lapa Romero, Rafael Garcia,
mining methods can be applicable to this problem as well. Hande Alemdar, and João LD Comba. 2019. How do soccer teams coordinate
Finally, there are several possibilities to enhance our shape consecutive passes? A visual analytics system for analysing the complexity of
passing sequences using soccer flow motifs. Computers & Graphics 84 (2019),
descriptor and glyph representation. These currently focus on 122–133.
the interacting trajectories between two players and the ball and [15] Rui Marcelino, Jaime Sampaio, Guy Amichay, Bruno Gonçalves, Iain D Couzin,
are agnostic to the surrounding context such as the positioning of and Máté Nagy. 2020. Collective movement analysis reveals coordination
tactics of team players in football matches. Chaos, Solitons & Fractals 138
nearby players, which however can influence individual interac- (2020), 109831.
tions to some degree. Including this information could therefore [16] Daniel Memmert, Koen APM Lemmink, and Jaime Sampaio. 2017. Current
add to the expressiveness and suitability to capture and explain approaches to tactical performance analyses in soccer using position data.
Sports Medicine 47, 1 (2017), 1–10.
individual interactions in a game. [17] Charles Perin, Romain Vuillemot, and Jean-Daniel Fekete. 2013. SoccerStories:
Ultimately, we are going to evaluate the utilization and usabil- A Kick-off for Visual Soccer Analysis. IEEE Trans. Vis. Comput. Graph. 19, 12
(2013), 2506–2515. https://doi.org/10.1109/TVCG.2013.192
ity of our system and its individual components, i.e., views on [18] Charles Perin, Romain Vuillemot, Charles D. Stolper, John T. Stasko, Jo Wood,
the data, together with end users such has soccer coaches. and Sheelagh Carpendale. 2018. State of the Art of Sports Data Visualization.
Comput. Graph. Forum 37, 3 (2018), 663–686. https://doi.org/10.1111/cgf.13447
[19] Telmo J. P. Pires. and Mário A. T. Figueiredo . 2017. Shape-based Trajectory
8 CONCLUSION Clustering. In Proceedings of the 6th International Conference on Pattern Recog-
nition Applications and Methods - Volume 1: ICPRAM,. INSTICC, SciTePress,
We have shown that the visual analysis of soccer games with 71–81. https://doi.org/10.5220/0006117400710081
a focus on the interactions between the players is a promising [20] Ángel Ric Diez, Carlota Torrents Martín, Bruno Gonçalves, Jaime Sampaio,
yet barely researched approach to game exploration. Although and Robert Hristovski. 2016. Soft-assembled multilevel dynamics of tactical
behaviors in soccer. Frontiers in Psychology, 2016, vol. 7, núm. 1513, p. 1-12
we demonstrated our approach only on the example of soccer (2016).
games, it should generalize to other team sports as well. We [21] Dominik Sacha, Feeras Al-Masoudi, Manuel Stein, Tobias Schreck, Daniel A
Keim, Gennady Andrienko, and Halldór Janetzko. 2017. Dynamic visual
believe that this approach can have a great potential in helping abstraction of soccer movement. In Computer Graphics Forum, Vol. 36. Wiley
coaches improve the performance of their teams based on the Online Library, 305–315.
deeper insight on their weaknesses and strengths. [22] Jaime Sampaio and Vitor Maçãs. 2012. Measuring tactical behaviour in football.
International journal of sports medicine 33, 05 (2012), 395–401.
[23] Roeland Scheepens, Christophe Hurter, Huub van de Wetering, and Jarke J.
REFERENCES van Wijk. 2016. Visualization, Selection, and Analysis of Traffic Flows. IEEE
Trans. Vis. Comput. Graph. 22, 1 (2016), 379–388. https://doi.org/10.1109/
[1] Gennady L. Andrienko, Natalia V. Andrienko, Peter Bak, Daniel A. Keim, TVCG.2015.2467112
and Stefan Wrobel. 2013. Visual Analytics of Movement. Springer. https: [24] Lin Shao, Dominik Sacha, Benjamin Neldner, Manuel Stein, and Tobias Schreck.
//doi.org/10.1007/978-3-642-37583-5 2016. Visual-interactive search for soccer trajectories to identify interesting
[2] Gennady L. Andrienko, Natalia V. Andrienko, Guido Budziak, Tatiana von game situations. Electronic Imaging 2016, 1 (2016), 1–10.
Landesberger, and Hendrik Weber. 2018. Exploring pressure in football. In [25] Aidan Slingsby and E. Emiel van Loon. 2016. Exploratory Visual Analysis
Proceedings of the 2018 International Conference on Advanced Visual Interfaces, for Animal Movement Ecology. Comput. Graph. Forum 35, 3 (2016), 471–480.
AVI 2018, Castiglione della Pescaia, Italy, May 29 - June 01, 2018. 54:1–54:3. https://doi.org/10.1111/cgf.12923
https://doi.org/10.1145/3206505.3206558 [26] Manuel Stein, Johannes Häußler, Dominik Jäckle, Halldór Janetzko, Tobias
[3] Natalia V. Andrienko and Gennady L. Andrienko. 2006. Exploratory analysis Schreck, and Daniel A Keim. 2015. Visual soccer analytics: Understanding the
of spatial and temporal data - a systematic approach. Springer. https://doi.org/ characteristics of collective team movement based on feature-driven analysis
10.1007/3-540-31190-4 and abstraction. ISPRS International Journal of Geo-Information 4, 4 (2015),
[4] Rita Borgo, Johannes Kehrer, David H. S. Chung, Eamonn Maguire, Robert S. 2159–2184.
Laramee, Helwig Hauser, Matthew Ward, and Min Chen. 2013. Glyph-based [27] Manuel Stein, Halldor Janetzko, Thorsten Breitkreutz, Daniel Seebacher, To-
Visualization: Foundations, Design Guidelines, Techniques and Applications. bias Schreck, Michael Grossniklaus, Iain D. Couzin, and Daniel A. Keim. 2016.
In 34th Annual Conference of the European Association for Computer Graphics, Director’s Cut: Analysis and Annotation of Soccer Matches. IEEE Computer
Eurographics 2013 - State of the Art Reports, Girona, Spain, May 6-10, 2013. Graphics and Applications 36, 5 (2016), 50–60. https://doi.org/10.1109/MCG.
39–63. https://doi.org/10.2312/conf/EG2013/stars/039-063 2016.102
[5] Juri Buchmüller, Dominik Jäckle, Eren Cakmak, Ulrik Brandes, and Daniel A. [28] Manuel Stein, Halldor Janetzko, Andreas Lamprecht, Thorsten Breitkreutz,
Keim. 2019. MotionRugs: Visualizing Collective Trends in Space and Time. Philipp Zimmermann, Bastian Goldlücke, Tobias Schreck, Gennady Andrienko,
IEEE Trans. Vis. Comput. Graph. 25, 1 (2019), 76–86. https://doi.org/10.1109/ Michael Grossniklaus, and Daniel A Keim. 2017. Bring it to the pitch: Combin-
TVCG.2018.2865049 ing video and movement data to enhance team sport analysis. IEEE transactions
[6] Eren Cakmak, Hanna Schäfer, Juri Buchmüller, Johannes Fuchs, Tobias on visualization and computer graphics 24, 1 (2017), 13–22.
Schreck, Alex Jordan, and D Keim. 2020. MotionGlyphs: Visual Abstrac- [29] Manuel Stein, Halldór Janetzko, Andreas Lamprecht, Daniel Seebacher, Tobias
tion of Spatio-Temporal Networks in Collective Animal Behavior. In Computer Schreck, Daniel Keim, and Michael Grossniklaus. 2016. From game events
Graphics Forum, Vol. 39. https://doi.org/10.1109/TVCG.2013.192 to team tactics: Visual analysis of dangerous situations in multi-match data.
[7] Somayeh Dodge, Patrick Laube, and Robert Weibel. 2012. Movement similarity In 2016 1st International Conference on Technology and Innovation in Sports,
assessment using symbolic representation of trajectories. Int. J. Geogr. Inf. Sci. Health and Wellbeing (TISHW). IEEE, 1–9.
26, 9 (2012), 1563–1588. https://doi.org/10.1080/13658816.2011.630003 [30] Tatiana von Landesberger, Sebastian Bremm, Tobias Schreck, and Dieter W.
[8] Joel Estephan, Joachim Gudmundsson, Michael Horton, and Sanjay Chawla. Fellner. 2014. Feature-based automatic identification of interesting data seg-
2017. Classification of Passes in Football Matches Using Spatiotemporal Data. ments in group movement data. Information Visualization 13, 3 (2014), 190–212.
(2017). https://doi.org/10.1177/1473871613477851
[9] Joachim Gudmundsson and Michael Horton. 2017. Spatio-temporal analysis [31] Yingcai Wu, Xiao Xie, Jiachen Wang, Dazhen Deng, Hongye Liang, Hui Zhang,
of team sports. ACM Computing Surveys (CSUR) 50, 2 (2017), 1–34. Shoubin Cheng, and Wei Chen. 2019. ForVizor: Visualizing Spatio-Temporal
[10] Hanqi Guo, Zuchao Wang, Bowen Yu, Huijing Zhao, and Xiaoru Yuan. 2011. Team Formations in Soccer. IEEE Trans. Vis. Comput. Graph. 25, 1 (2019), 65–75.
Tripvista: Triple perspective visual trajectory analytics and its application on https://doi.org/10.1109/TVCG.2018.2865041
microscopic traffic data at a road intersection. In 2011 IEEE Pacific Visualization
Symposium. IEEE, 163–170.