Extraction of goal-scoring segments and player positions in a panoramic ice hockey video

This paper presents a method for extracting goal-scoring segments and player positions from a panoramic ice hockey video. The proposed approach extracts the goal-scoring segments by detecting the goal light brightness changes and analyzing the player distribution. The extracted segments are processed with multiple player detection and tracking methods to acquire the player positions. The proposed method employs Faster R-CNN and YOLO11 for the player detection, and tracking methods such as Segment-Anything-2-Real-Time and OCSORT for the player tracking. In the experiments, the processing time and tracking accuracy of the proposed method are evaluated.