Classifying Corner Kicks in Football
Authors
Gregersen, Jakob Faarbæk ; Nesheim, Alexander ; Leth, Casper Kjærhus
Term
4. term
Education
Publication year
2024
Submitted on
2024-05-24
Pages
14
Abstract
Small edges can decide a football match, and set pieces are often pivotal. This study uses machine learning to classify corner kicks to support analysts and coaches. We use tracking data of player and ball positions recorded at 25 frames per second from 2,132 corner-kick situations. From these data, we derive eight features that describe each situation and test how well they classify corner types, both individually and in combination. Drawing on a coaching guide, we define six types of corner kicks and label the dataset accordingly. We also examine the effect of upsampling—that is, increasing the number of examples of rarer types to reduce class imbalance. Our findings indicate that these features and their combinations provide a strong starting point for future research on automatic corner-kick classification.
Små fordele kan afgøre en fodboldkamp, og standardsituationer er ofte nøglen. Denne undersøgelse bruger maskinlæring til at klassificere hjørnespark for at give analytikere og trænere brugbare indsigter. Vi anvender positionsdata for spillere og bold, registreret med 25 billeder pr. sekund, fra 2.132 hjørnesparkssekvenser. Ud fra disse data udleder vi otte egenskaber (features), som beskriver situationen, og vi vurderer deres evne til at skelne mellem klasser – hver for sig og i kombination. Med udgangspunkt i en trænerguide definerer vi seks typer hjørnespark og annoterer datasættet derefter. Vi undersøger også effekten af opsampling, dvs. at øge antallet af eksempler i sjældne klasser for at udligne skævheder i datasættet. Vores analyser peger på, at de valgte egenskaber og deres kombinationer giver et solidt afsæt for fremtidig forskning i automatisk klassificering af hjørnespark.
[This apstract has been rewritten with the help of AI based on the project's original abstract]