The Barça Sports Analytics Summit is looking for the most relevant and innovative contributions in the field of football analytics. For the 2019 edition, we chose 8 papers from all the papers received, for the authors to present their findings to leading football analytics representatives.
Using contextual player movement and spatial control to analyse player passing trends in football
Measuring a team’s spatial control using motion models fit on average behaviour underestimates players’ ability to perform high effort movements (e.g., reorientation). This paper presents a method for producing probabilistic motion models with consideration of movement context. Models were fit on sampled movement from the 2018 MLS season. Passing risk was estimated via the attacking team’s spatial control and used to produce player passing networks. This analysis has applications in player profiling, tactics, and recruitment.
Dynamic analysis of team strategy in professional football
Team formations are the foundations of tactics in football. However, in the modern game, formations are far more dynamic and dependent on game state than ‘4-4-2’ or ‘3-5-2’ imply. In this paper, a large sample of tracking data is used to measure how team formations vary throughout a match. Transitions from defence to attack are analysed, and the impact of major tactical changes on match outcomes are studied.
Estimating Locomotor Demands During Team Play from Broadcast-Derived Tracking Data
Optical tracking data permits sports scientists to estimate external load metrics used to understand the physical toll a game takes on an athlete. Unfortunately, tracking data is not widely available. Computer vision techniques allow broadcast video to be converted to coordinates, making tracking data easier to acquire, but data is censored when players are offscreen. In this research, models are developed that predict offscreen load metrics and demonstrate the viability of broadcast-derived tracking data for understanding external load in soccer.
Ready Player Run: Off-ball run identification and classification
One of the major downfalls of tracking data in football is the lack of a common language to describe actions that take place off the ball, particularly patterns of player movement. This paper provides a method for identifying and classifying off-ball in possession runs into similar groups to allow for more generalisable analysis. The objective is to create a vocabulary of run types that can be used to better describe or analyse specific runs and be queried more easily than raw tracking data.
Automating insight extraction from football data visualizations
This paper introduces similarity metrics to create heatmaps and passing sonars, as well as illustrating how to use them to find players that are particularly similar or dissimilar in their moves or passing intentions. A visual tool is also explained which represents the average player’s passing intentions given a fixed heatmap and uses it to classify players by their passing predictability. Finally, a framework to automate insight extraction from variations in a player’s heatmap or passing sonar with time or game circumstances.
Landscapes of passing opportunities in Football – where they are and for how long are available?
Using players positional data from a competitive football match, landscapes of passing opportunities were created and categorized into three groups of passes: i) penetrative, ii) support and iii) backward. Displayed as heatmaps these landscapes show more passing opportunities on the second half. Furthermore, results display that penetrative passes were available for shorter periods than backward passes that were available for shorter periods than support passes. This customizable tool provides insights into attacking dynamics allowing collective and individual player performance analysis.
Luis Gómez-Jordana Martin
Explainable Injury Forecasting in Soccer via Multivariate
Time Series and Convolutional Neural Networks
Injuries have a significant impact on professional football due to their influence on performance and the considerable costs of rehabilitation for players. By exploiting an electronic performance tracking system, we can represent a player’s workload history as a Multivariate Time Series (MTS). This MTS can be used to train a Convolutional Neural Network that forecasts whether or not a player will get injured in a future time window. This injury forecaster is both accurate and explainable, allowing a club’s staff to interpret the reason behind a player’s injury easily.
Head, Shoulders, Hip and Ball… Hip and Ball! Using Pose Data to Leverage Football Player Orientation
Orientation has proven to be a key skill for football players in order to succeed in a broad spectrum of plays. However, body orientation is a yet a little-explored area in sports analytics. By seeking the 2D orientation of the field projection of a normal vector placed in the center of the upper torso of players, this research presents a novel technique to extract orientation automatically from video recordings by merging pose and contextual information. Results have been validated with players using an EPTS device, obtaining a median error of fewer than 35 degrees/player.