Individual action and group activity recognition in soccer videos. Master's thesis
Data and statistics are key to soccer analytics and have important roles in player evaluation and fan engagement. Automatic recognition of soccer events - such as passes and corners - would ease the data gathering process, potentially opening up the market for non-professional soccer analytics. We propose a novel method for the automatic recognition of soccer events from video. To the best of our knowledge, it is the first method that infers both individual actions and group activities simultaneously from soccer videos. Three key contributions in the proposed method are (1) the use of player-centric snippets as model input, (2) per-player feature extraction with an I3D CNN - based on RGB video and optical flow - and (3) the use of feature suppression and zero-padding in graph attention networks for feature contextualisation. The results show that the proposed method performs better than an alternative state-of-the-art method, designed for action and activity recognition in volleyball. Our method gains 98.7% accuracy for the recognition of eight actions and 75.2% for eleven activities.
To reference this document use:
Group activity recognition
Soccer match events
University of Twente
begeleiding vanuit TNO: H.Dol W.R. Uijens