[4Rin1] Interactive 2

[4Rin1-33] Suggestion of Strategy Framework for Ice-Hockey Using Videos

〇Takuya Shimano, Ryota Nagano¹, Shuya Bundo, Yusaku Mandai (1.HockeyInnovation Inc)

Keywords:sports analytics, ice hockey match data, video processing, valuing actions, probabilistic classification

In recent years, the NHL, the world's largest ice hockey league, is actively conducting strategic analysis using game videos. However, most services use high-performance and expensive cameras and sensors, and therefore it is difficult for amateur teams to conduct that kind of analysis. This paper proposes a new framework that analyzes strategies automatically from only video shooting. We focus on the following topics: (1) Object detection (e.g., player, goal and face-off) using Mask-RCNN. (2) Homography transformation for estimation of players’ coordinates on hockey rink from positional relationship of objects. (3) Estimation of the expected goal values from players' coordinates and actions using a probabilistic classifier. We verified the accuracy of those using videos from NHL 2019-20. As a result, although some issues (e.g., false-positive, expected goal value accuracy) have to be resolved, we demonstrated that our method was sufficiently possible.