Recognition of basketball referee signals using machine learning and computer vision
View/
Keywords
Computer vision ; Deep learning ; Basketball referee signal recognition ; Gesture recognition ; YOLOv8 ; Video analysisAbstract
This thesis investigates the problem of automatic basketball referee signal recognition from
broadcast video footage. Unlike conventional object recognition tasks, referee signals are expressed
through subtle variations in human pose and hands configuration, making them challenging to detect
under realistic broadcast conditions characterized by camera motion, resolution, occlusions, scale
variation, and contextual bias. To address these challenges, a referee centric, multistage computer
vision pipeline is proposed. The system decomposes the task into three sequential components. First,
referee detection from full broadcast frames. Second, binary classification to distinguish signal from
non-signal gestures and third, multi-class classification to recognize specific signal types. The approach
is evaluated against a direct broadcast frame classification baseline, demonstrating that context driven
models achieve high apparent accuracy but fail to generalize reliably. Custom datasets were
constructed from full game recordings under realistic conditions. Experimental results show that
detection-first architectures improve robustness, while dataset scale, class balance, and augmentation
strategy critically influence generalization performance. Furthermore, lightweight temporal
aggregation mechanisms are integrated to enhance prediction stability during video inference without
requiring computationally intensive end to end video models. The findings highlight the importance of
architectural decomposition and domain dataset construction in gesture recognition systems
operating in real world broadcast environments. The proposed framework provides both
methodological insights and practical guidelines for developing structured recognition systems in
sports analytics.
Postgraduate Studies Programme
Πληροφοριακά Συστήματα και ΥπηρεσίεςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνNumber of pages
89Language
EnglishCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα