Collaborative Multi-Camera Tracking of Athletes in Team Sports
Wei Du, Jean-Bernard Hayet, Justus Piater, Jacques Verly · ORBi (University of Liège) · 2006
Abstract. A novel approach to tracking athletes in team sports using multiple cameras is proposed that addresses several issues including occlusions and propagation of wrong information. The strength of this approach lies in the use of belief propagation which enables good observations in some views to compensate for poor observations in other views due e.g. to occlusions. Each target is tracked in each view by a dedicated particle-filter-based local tracker. The trackers in different views interact with each other via belief propagation so that a local tracker operating in one view is able to take advantage of additional information from other views. By combining particle filters and belief propagation in a unified framework, we develop a sequential belief propagation algorithm to perform inference of multi-view target states collaboratively. We demonstrate the effectiveness of our method on sequences of soccer games. 1