SSET: a dataset for shot segmentation, event detection, player tracking in soccer videos

Abstract

Soccer video analysis is the focus of sports video research as it receives widespread attention around the world. However, the lack of soccer datasets hinders the rapid development of this field. In this paper, we construct a soccer dataset named Soccer Dataset for Shot, Event, and Tracking (SSET), which can meet the research needs of shot segmentation, soccer event detection and player tracking. So far, we have collected 350 soccer videos, involving a variety of soccer games, for a total of 282 h. The dataset consists of three parts: (1) Shot, including five shot types and two shot transition types; (2) Event/Story, consisting of 11 fine-grained event and 15 coarse-grained story types where the story extends the event types with 4 extra types; (3) Bounding box of players, giving the coordinates, width and length of the bounding box. In addition, we develop an annotation tool called Sports Video Dataset Markup (SVDM) for sports video data annotation and hope that more people join our work. We conduct event detection and player tracking experiments on our dataset, and the results show that the existing works are not completely suitable for solving soccer video analysis tasks. Our dataset is available at http://media.hust.edu.cn/dataset.htm .