Tracking of Multiple Targets Using Online Learning for Reference Model Adaptation

Recently, much work has been done in multiple ob-4 ject tracking on the one hand and on reference model adaptation5 for a single-object tracker on the other side. In this paper, we do6 both tracking of multiple objects (faces of people) in a meeting7 scenario and online learning to incrementally update the models8 of the tracked objects to account for appearance changes during9 tracking. Additionally, we automatically initialize and terminate10 tracking of individual objects based on low-level features, i.e., face11 color, face size, and object movement. Many methods unlike our12 approach assume that the target region has been initialized by13 hand in the first frame. For tracking, a particle filter is incor-14 porated to propagate sample distributions over time. We discuss15 the close relationship between our implemented tracker based16 on particle filters and genetic algorithms. Numerous experiments17 on meeting data demonstrate the capabilities of our tracking18 approach. Additional...

• Type: Journal
• Year: 2008
• Where: TSMC
• Authors: Franz Pernkopf
• Keywords: Multiple Ob-4 Ject, Particle Filters, Reference Model, TSMC 2008