Skip to main navigation menu Skip to main content Skip to site footer

Vol. 11 No. 1 (2025)
Special Issue: Proceedings of CVIS 2025

Articles

Temporally Stable Rink Homography Estimation via 3D Reconstruction and Segmentation Fusion

https://doi.org/10.15353/jcvis.v11i1.10020

Published 2026-04-01

  • Liam Salass
    • ,
  • Bowen Dai
    • ,
  • Yuhao Chen
    • ,
  • David Clausi
    • ,
  • John Zelek

Abstract

Homography estimation in broadcast hockey video is challenging due to frequent occlusions, motion blur, and limited visibility of rink markings. Prior work estimates a homography independently for each frame, leading to substantial temporal jitter that undermines downstream tasks such as trajectory analysis and event recognition. We introduce a two-stage homography pipeline that leverages sequence-level 3D reconstruction and aggregated segmentation evidence. A stable rink plane is first extracted from a monocular 3D reconstruction, after which per-frame rink segmentations are warped into the plane and fused to recover the rink’s global layout. A single plane-to-template homography is then estimated, producing a temporally stable and metrically consistent mapping from image space to rink coordinates. Experiments show that puck trajectories transformed using our homography are smooth and free of frame-to-frame jitter.
PDF