Soccer player tracking using particle filters

Abstract

We present a multi-target tracking system for estimating the position of multiple soccer players as they move around during a soccer game. Our system relies on silhouette observations recorded by a static camera, and utilises 3D models and particle filter methods to estimate athlete positions. Tracking parameters can be adapted in order to tip the scales towards precision or performance. We present the system's software architecture as a client-server application, and demonstrate results based on a simulated yet realistic dataset produced using the Unity3D game engine.