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

VIP-HTD : A Public Benchmark for Multi-Player Tracking in Ice Hockey

Abstract

Multi-Object Tracking (MOT) is the combined task of localization and association of subjects across a temporal sequence. Unlike the popular pedestrian tracking paradigms, monocular tracking of ice hockey players from broadcast feeds presents a variety of challenges due to rapid non-linear motions, occlusions, blurs, and pan-tilt-zoom effects. To tackle these issues, there neither exists public datasets nor benchmarks trained on public datasets to date. To this end, we propose: (a) VIP-HTD - a public ice hockey tracking dataset, processed \& curated from existing work, and (b) a public benchmark for multi-player tracking based on it. Further, we also present our observations processing this dataset and discuss the two key metrics (IDF1 score and ID switches) required for optimal tracking evaluations. With this work, we take a step towards creating a unified public benchmark for evaluating multi-player tracking in hockey.
PDF