This study presents a novel approach to improve multi-object tracking within the field of Ice Hockey Analytics. By harnessing depth estimation, the goal is to tackle the common challenges related to tracking multiple objects in a 3D scene that is projected onto a 2D screen. The methodology encompasses acquiring video sequences, performing depth estimation for all frames, and subsequently conducting multi-object tracking based on the resulting depth images. While this extended abstract does not offer a comprehensive set of results, it does provide valuable qualitative insights.
PDF