logo SBA

ETD

Archivio digitale delle tesi discusse presso l’Università di Pisa

Tesi etd-08312022-180614


Tipo di tesi
Tesi di laurea magistrale
Autore
BALDI, FEDERICA
URN
etd-08312022-180614
Titolo
Action Detection in Rugby Video Sequences in both Professional and Amateur Settings
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
ARTIFICIAL INTELLIGENCE AND DATA ENGINEERING
Relatori
relatore Prof. Falchi, Fabrizio
relatore Prof. Gennaro, Claudio
relatore Dott. Marchesotti, Luca
Parole chiave
  • action detection
  • computer vision
  • deep learning
  • video understanding
Data inizio appello
23/09/2022
Consultabilità
Completa
Riassunto
Among the several problems that are part of Computer Vision and, more specifically, Video Understanding, we find action detection, i.e., the task of classifying and temporally locating actions that appear in a video. This is a widespread field of research at the moment, probably also because of the countless real-world applications that could benefit from it.
In recent years and owing mainly to the advent of new Deep Learning technologies, exceptional progress has been made. However, as it is currently almost utopian to be able to devise a generic architecture that works well regardless of the context, there is still a lot of research to be done in the different application areas. One of the most popular yet challenging one is the analysis of sports videos.
Within this context, the objective of this thesis is to address the problem of automatic action detection in rugby video sequences, filmed in both professional and amateur settings. So far, and to the best of our knowledge, research in the specific use case of rugby is lagging far behind that of other sports, and there is neither a Machine Learning method nor datasets published for this purpose.
Therefore, first, a dataset was created by collecting and annotating rugby matches at various levels of professionalism. The intent is to make the dataset publicly available so as to jumpstart research in this area as well. Then, and after careful inspection of the state of the art, experiments on the new dataset were performed by adapting one of the existing approaches in the literature. The approach, originally designed for soccer, allowed us to create a baseline on the new dataset and evaluate the differences between the performance obtained on the professional segment and that obtained on the amateur one. Finally, after analyzing the results, possible future lines of research were proposed.
File