Action Spotting in Soccer broadcast Untrimmed videos
Published:
In team sports like football both person level information and group level contexts are crucial for event recognition. The task of spotting consists of finding the anchor time that identifies an event in a video and then classify the event to appropriate class. A common pipeline comprise of proposing temporal segments which are in turn further pruned and classified. Common methods for activity classification and detection make use of dense trajectories, actions’ estimation, recurrent neural network, tubelets and handcrafted features. In order to recognize or detect activities within a video, a common practice consists of aggregating local features and pooling them, looking for a consensus of characteristics. In this project we focus our analysis on action spotting in soccer broadcast videos.