Toward Automatically Labeling Situations in Soccer

Abstract

We study the automatic annotation of situations in soccer games. At first sight, this translates nicely into a standard supervised learning problem. However, in a fully supervised setting, predictive accuracies are supposed to correlate positively with the amount of labeled situations: more labeled training data simply promise better performance. Unfortunately, non-trivially annotated situations in soccer games are scarce, expensive and almost always require human experts; a fully supervised approach appears infeasible. Hence, we split the problem into two parts and learn (i) a meaningful feature representation using variational autoencoders on unlabeled data at large scales and (ii) a large-margin classifier acting in this feature space but utilize only a few (manually) annotated examples of the situation of interest. We propose four different architectures of the variational autoencoder and empirically study the detection of corner kicks, crosses and counterattacks. We observe high predictive accuracies above 90% AUC irrespectively of the task.

Putting team formations in association football into context

“…The main objective of this paper was to detect phases of play as a preliminary for contextualized formation analysis. Previous work has attempted to detect only single specific phases of play, such as counterattacking (Fassmeyer et al, 2021;Hobbs et al, 2018) or counterpressing (Bauer, 2021;Bauer et al, 2021). For the first time, we present a method for classifying games into five distinct phases of play.…”

“…These defensive phases were also sample-data). Those that do focus on finding a single specific transition phases, such as counterattacking (Decroos et al, 2018;Fassmeyer et al, 2021;Hobbs et al, 2018) or using the raw positional data as input instead of Table 2 shows some basic statistics for the training data, including the F 1 -score Mid-block and build-up are clearly the dominant phases, making up 39% and 47% of the phases shown in Table 2. They are also the phases with the longest duration, lasting an average of 19.0 seconds (midblock) and 18.6 seconds (build-up).…”

“…The complexity of a football match requires so called team tactics in which the whole team is involved (Rein et al, 2016). Some exemplary patterns like counterattacks (Fassmeyer et al, 2021;Hobbs et al, 2018), ball regain strategies (Vogelbein et al, 2014), i.e. counterpressing (Bauer et al, 2021) or general offensive strategies (Decroos et al, 2018;Grunz et al, 2012; The project therefore combines machine learning and human experience aiming to obtain results that are insightful, meaningful and of practical use to coaches, managers, and scouts.…”

Ball Trajectory Inference from Multi-Agent Sports Contexts Using Set Transformer and Hierarchical Bi-LSTM

“…Sorano et al [44] studied the problem of detecting passes from soccer videos by utilizing a CNN-based object detection engine (YOLOv3) and Bi-LSTM. Fassmeyer et al [17] proposed a method that detects a wider range of events such as corner kicks, crosses, and counterattacks using a Variational Autoencoder (VAE) and Support Vector Machine (SVM). Note that all of these approaches rely on the ball tracking information from video data [44,49] or manual annotation [17].…”