This line chart, reflects the different changes of pace for one or more players in the given analysis interval. The analyst can browse through the data evaluating the speed progression in the selected time interval. In line with other visualizations presented in this paper, this chart is also interactive in the way that it allows the researcher to navigate to the detected key event in the virtual player, effectively reducing the cognitive load that is involved in the analysis process.
We employ a stacked bars chart to allow fast visual comparison of the distance covered by players in a time interval. Each stack represents the portion of the total distance performed at each one of the different speed intervals presented above. This enables the analyst to have a bird's eye perspective of the effort performed by the different players in a group, giving the analyst the perception about the time periods in which a player performed at a certain pace.
Spatial distribution maps or heat maps are a popular means to visualize spatio-temporal variables such as position or speed. As a companion visualization for the stacked bars chart, the analyst also has the possibility to generate this kind of visualization on demand.
By employing heatmaps we are able to provide a better picture of a player's performance through any given period of time. Finally, the pilot tool allows analysts to make in-map measures among one or more players to a specific coordinate point in the field, as shown in Figure 3. A real elite soccer match was examined to demonstrate the capabilities of the proposed pilot tool. Firstly, a significative event—in this case a goal-, which is known by the user prior the analysis task, is presented.
Then the prototype is used to search for other secondary events within the relevant play build-up that support a certain hypothesis that explains the final outcome of the play. The other example proposes a free exploration of the dataset. By employing phase analysis, the analyst is able to identify significative events that generate different kinds of situations of interest in the game. As an operational example, Figure 4 includes the original video footage and the two-dimensional visual representation as seen in the tool of three different moments of a scoring play.
The key players intervening in the play are three attackers of the yellow team and three defenders of the red team. Covered area by the two groups is shown in the tool.
At the center of each area there is a dot filled with the color of the team. This allows us to know where the centroid of the three players is, allowing the user to see their average positions at all times. It is known there is an increase in the chances of scoring a goal when the attacking group's centroid surpasses that of the defenders.
In this use case, we verify how this technical event can be easily detected in our pilot tool to explain a real game event. Figure 4. Build-up phases of a scored goal. The area covered by attackers and defenders is depicted in the system. The first moment depicted in Figure 4 shows the start of the build-up, where Player 15 red gets the ball in the offense phase.
In the second moment, he has surpassed a midfielder and is approached by two of the three highlighted defenders, which creates an opportunity for Player 7 red to get a scoring chance. In the last moment, Player 15 has already passed the ball to Player 7, who will finish easily and score a goal due to the defenders being stuck in more advanced positions, making it impossible to catch him.
The second moment represents the key to the scoring chance and ultimately the goal.
By exploring the positioning data, we better understand the impact of the two defenders 20 and 5, yellow on the play. They decide to close onto the attacker with the ball 15, red and that causes the centroids of both groups to get at practically the same position. What is more, this triggers an unbalanced situation in which players 9 and 7 of the attacking team red are left alone against Player 3 of the defending one yellow.
When the two defenders that tried to prevent the pass, start recovering and running back, it is late enough for Player 7 red to be on a clear scoring position, as depicted in the third moment. The difference between both centroids at this point is the largest of the sequence. This fact is in line with the proposals from other authors previously explained Duarte et al.
Further explanatory facts can be found using individual analysis, for example if the speed evolution graphs of two of the players are considered: Player 7 attacker, red and Player 3 defender, yellow. As depicted in Figure 2 , Player 7 maintained a higher speed over Player 3 during the 8 analyzed seconds.
The most important moments stand between the and marks, as Player 7 sprinted toward the goal a second after the defender started his movement. This allowed him to gain sufficient advantage to be alone in a scoring position later, as seen in the last moment depicted in Figure 5. Figure 5. Speed comparison between players 7 attacking and 3 defending during the analyzed play.
Relative phase can also be used to identify inter-player coordination. Two instants between minutes 28 and 38 are highlighted in Figure 6 depicting the times when the coordination between Player 3 and 6 of the red team defending was low.
This implies that one defender moved toward the centroid while the other moved away from it, creating an unbalanced situation in the defense which has important implications on the outcome of the play for example the first instant corresponds to a goal score by the opposing team. Figure 6. Phase analysis for the two defenders between minutes 26 and 38 of the second half. The two lowest moments of coordination are rounded in the image. The two instants of the sequence are depicted in Figure 7. By reproducing the play, we can see how the two defenders go back to cover Player 8 yellow as he surpasses them, trying to prevent a scoring chance.
This attacking player then passes the ball back and that causes the defenders to lose their coordination, as one stays still at the goal line and the other runs to prevent the shot from the attacking player number 15, yellow who is outside of the goalkeeper's area.
Moments of low coordination between defenders of the red team are generated in situations where the defenders are close together and an unexpected action such as a back pass happens. This unpredictable situation forces the defenders to lose synchronization with the rest of his teammates meaning each one of them took different directions, as it happened in the presented situation: In this concrete case, one decided to stay put in the goal line and the other to cover the shot, thus creating completely opposite movements from the centroid of the whole defense.
Figure 7. First moment of low coordination between two defenders Players 3 and 6, red.
Mark Dennison. Steve Capell. Stelian Coros. Jose A. These data sets are nowadays obtained using a variety of techniques, including multi-view stereo reconstruction methods from multiple 2D images.
The outcome is a goal by the attacking team. The second point of low coordination is depicted in Figure 8. In this case, the play did not result in a scoring chance, although we comment on other interesting aspects of the game that generated this situation. In this play, player 5 passes the ball to Player 20 both attacking, in yellow , and Player 30 defending, red is not fast enough to stop the attack, making Player 6 red to leave its defense position to cover the area left by his team mate. It becomes clear that bad defense by Player 30 caused a break in the coordination of both central defenders 3 and 6, red.
While Player 3 had to remain covering the attacker Player 9, yellow , Player 6 had to go out of his zone of influence to stop Player 20 and prevent a cross or any other potentially dangerous play on offense. This special situation could also be easily identified in the pilot tool by employing the line graph, without the need to visualize the entire 6 min of game. Figure 8. Top: Second moment of low coordination between two defenders players 3 and 6, red. Bad defending from another player causes the disruption. Bottom: Speed and movement comparison between the two defenders involved in the opposition advantage play.
In this study, we demonstrate that including visual analysis techniques in sports group behavior analysis can produce both high quality results and a satisfactory experience with the computer at the same time see Supplementary Video 1.
We designed an iterative workflow companion to this tool that effectively captures and accelerates many expert workflows that had to be performed manually or semi-automatically in the past. We found that the application of data visualization and user-centered design techniques helps to maintain the overall learning curve of the system flat without compromising the accuracy or efficiency of the analysis.
In the provided examples, we could also appreciate how the proposed workflow can scale to many different game situations and combinations of analysis types -collective or individual. By following a top-down approach that goes from the general to the specific, the analyst can navigate the positional data by progressively tuning the interface controls and reach to conclusions of interest. Furthermore, the advantages of combining the proposed workflow with a visual tool could be noticed in the implementation of two different use cases that employed geospatial data acquired from a professional match in a very similar setting to what many sports scientists have in their daily work.
Despite of these promising first results, the ideas presented in this paper require of further validation and more complex user studies that we expect to implement in future research, along with the following lines of future work that we compiled during our research:. As it has been commented before in this paper, some authors take the ball position into account when calculating different group metrics.
Given the recent advances in positioning technologies, we expect to be able to work with more data sets that include this feature. As a result of the incorporation of the ball position, we plan to generate social network visualizations that depict associations of players during the game, extracting statistical data related to the number of passes between players, direction of the pass and other traits, which will be used as inputs for our system.
This information should be presented to the user by employing dynamic network analysis and other complex network visualization techniques. In our current implementation, we discretize the speed variable according to five predefined clusters. In a similar way, we could employ other non-geospatial variables gathered using different equipment attached to the athlete's body, such as a pulsometer. Different studies divide the effective playing space of a group of players in several areas of influence one per player.
There are previous attempts to achieve this functionality in visual terms by making use of Delaunay triangulations and Voronoi cells. The pilot tool is able to link incoming metrics data with real in-game situations by establishing an interactive relationship between the virtual player and the two analysis zones. During analysis sessions, analysts tend to complement the usage of these prototypes with real video playback of the game under study. Currently this process must be performed manually, but it would be interesting to reduce the number of steps needed to reach to relevant parts of the video in a similar way to how the virtual player performs in the current tool.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Supplementary Video 1.