Faceoffs are pivotal events in hockey, marking
strategic resets in gameplay that influence team
positioning and puck possession. Detecting these
events in video data can provide valuable insights
for coaches and analysts, enabling the study of
player formations and strategies around these critical
moments. This work presents a novel framework
for detecting ice hockey faceoffs. Our approach
processes overhead video sequences using
a multi-stage pipeline, incorporating object detection
and segmentation to track player trajectories
across frames. We employ state of the art detectors
and tracking tools, enabling tracking and trajectory
analysis for each player. Additionally, preprocessed
sequences are used to better ensure accurate
player tracking. We demonstrate the framework’s
effectiveness in automatically identifying faceoffs,
with promising results that suggest its potential for
broader applications in sports analytics. By enhancing
the visibility of faceoffs and player interactions
in hockey, this work contributes toward automated
sports analytics, providing a robust tool for studying
patterns and tactics in high-paced, dynamic sports
environments.
PDF