Object tracking based on kernel recursive least-squares with total error rate minimization

This paper presents an online tracking system which considers both target appearance and background changes simultaneously. Based on a kernel technique, a recursive formulation is proposed for total-error-rate (TER) minimization. Subsequently, the online solution is integrated into particle filtering to effectively distinguish the target object from the background. Our system is compared qualitatively and quantitatively with related existing methods on publicly available video sequences.